⚡ electrical engineering

Semi-Blind Joint Channel and Symbol Estimation for Beyond Diagonal Reconfigurable Surfaces

This paper proposes a semi-blind tensor-based approach for joint channel and symbol estimation in Beyond Diagonal Reconfigurable Intelligent Surfaces (BD-RIS) under time-varying mobility conditions, introducing two novel receivers based on PARATUCK-to-PARAFAC transformation and TUCKER decomposition that eliminate the need for pilot sequences while demonstrating superior performance over existing methods.

Original authors: Gilderlan Tavares de Araújo, André L. F. de Almeida, Buno Sokal, Gabor Fodor

Published 2026-03-17

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Gilderlan Tavares de Araújo, André L. F. de Almeida, Buno Sokal, Gabor Fodor

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: The "Magic Mirror" Problem

Imagine a future where wireless signals (like your Wi-Fi or 5G) can be bounced off special walls or surfaces called RIS (Reconfigurable Intelligent Surfaces). Think of these surfaces as "Magic Mirrors" that can catch a signal from a cell tower and reflect it perfectly to your phone, even if you are in a dead zone behind a building.

For a long time, these mirrors were "single-connected." This means every little piece of the mirror (let's call them tiles) could only reflect the signal in one specific way, independently of its neighbors. It was like a choir where every singer could only sing one note, and they couldn't talk to each other.

The New Tech: BD-RIS
This paper introduces a new, smarter mirror called BD-RIS (Beyond Diagonal RIS). In this version, the tiles are all connected to each other. They can "talk" and coordinate. If Tile A reflects a signal, it can influence how Tile B reflects it. This creates a much more powerful, flexible signal, like a choir where everyone harmonizes perfectly.

The Problem: How do we tune the mirror?
To make this magic work, the cell tower needs to know exactly how the signal travels from the tower to the mirror, and then from the mirror to your phone. This is called Channel Estimation.

Usually, to figure this out, the tower has to send a special "test song" (called a pilot sequence) before sending your actual data (like a video call).

The Downside: Sending a test song takes up time and bandwidth. It's like a radio station playing a 5-minute commercial before every song. It slows everything down and wastes energy.

The Paper's Solution: "Semi-Blind" Listening

The authors of this paper say: "Why send a test song at all? Let's just listen to the actual music you're trying to send."

They propose a Semi-Blind approach. Instead of stopping to send a test signal, the system uses the actual data you are sending (your voice, your video) to figure out how the mirror is behaving while it's happening. It's like a sound engineer who can tune the acoustics of a room just by listening to the band play, without ever asking them to stop and play a single note.

The Secret Weapon: "Tensor" Math

To do this, the authors use a fancy math tool called Tensor Decomposition.

The Analogy: Imagine you have a giant, 4-layered cake (the signal).
- Layer 1: Time (when the signal arrives).
- Layer 2: Space (which antenna caught it).
- Layer 3: The Mirror's settings (how it was tuned).
- Layer 4: The Data (your message).

In the past, engineers tried to slice this cake into 2D layers (like cutting a sandwich) to understand it. This was messy and lost information.
The authors realized that if you look at the cake as a whole 4D object, you can use a special mathematical "knife" (Tensor Decomposition) to slice it perfectly, separating the "Time" from the "Space" and the "Mirror Settings" from the "Data" simultaneously.

The Two New "Recipes" (Algorithms)

The paper proposes two different ways to slice this 4D cake. They call them PAKRON and TUCKER.

1. The PAKRON Recipe (The Two-Step Dance)

How it works: This method takes a two-step approach.
- Step 1: It guesses the general shape of the mirror's behavior and the data combined.
- Step 2: It takes that guess and tries to separate the "Mirror" part from the "Data" part.
Pros: It's faster and uses less computer power. It's like a quick, efficient dance.
Cons: Because it does it in two steps, if it makes a small mistake in Step 1, that mistake gets carried over to Step 2. It's like trying to clean a window; if you wipe it with a dirty cloth in step one, the second wipe won't be perfect.

2. The TUCKER Recipe (The All-in-One Masterpiece)

How it works: This method looks at the whole 4D cake at once. It doesn't separate the steps; it solves for the Mirror, the Data, and the Signal all at the same time in a single, complex loop.
Pros: It is much more accurate. It doesn't make the "dirty cloth" mistake because it sees the whole picture. It handles noise (static) much better.
Cons: It requires a supercomputer brain. It is much heavier on processing power and takes longer to calculate.

The Results: Who Wins?

The authors ran thousands of simulations (like running a million virtual parties) to see which method worked best.

Accuracy: The TUCKER method was the clear winner. It figured out the mirror settings and the data more accurately, especially when the signal was weak or noisy.
Speed: The PAKRON method was faster and used less energy.
The Big Win: Both methods were much better than the old way of sending "test songs" (Pilot-Assisted methods). By using the actual data to tune the mirror, they saved a huge amount of time and bandwidth.

The Takeaway

This paper is about teaching a smart, connected mirror (BD-RIS) how to tune itself using the actual conversation happening on it, rather than stopping to run a test.

If you have a powerful computer and need the best possible quality, use the TUCKER method.
If you need something fast and efficient and can tolerate a tiny bit of error, use the PAKRON method.

Both methods are a huge step forward because they stop wasting time on "test signals," making our future wireless networks faster, more efficient, and capable of connecting us even in the hardest-to-reach places.

1. Problem Statement

The paper addresses the challenge of channel estimation and data symbol detection in Beyond-Diagonal Reconfigurable Intelligent Surfaces (BD-RIS) assisted MIMO systems.

Context: Unlike traditional RISs where elements are single-connected (diagonal phase shift matrix), BD-RIS interconnects elements (e.g., group-connected or fully-connected) to create a non-diagonal scattering matrix. This enhances degrees of freedom but complicates channel estimation due to the increased number of coupling coefficients and the lack of signal processing capabilities at the RIS.
Limitations of Existing Methods:
- Pilot-Assisted Methods: Traditional approaches (e.g., Least Squares) require dedicated training sequences, leading to high pilot overhead and reduced spectral efficiency.
- Tensor Methods: Existing tensor-based solutions for BD-RIS often rely on pilots or separate channel estimation and data decoding stages.
Specific Challenge: The paper considers a practical scenario with time-varying User Terminal (UT)-to-RIS channels (due to mobility) while the RIS-to-Base Station (BS) channel remains static. The goal is to perform joint channel and symbol estimation without a dedicated pilot training phase to maximize spectral efficiency and reduce latency.

2. Methodology

The authors propose a data-driven, semi-blind approach that leverages the algebraic structure of the received signal using Tensor Decomposition.

A. System Model & Signal Structure

Architecture: A single-user uplink MIMO system with a group-connected BD-RIS.
Scattering Matrix Design: The BD-RIS scattering matrix $S_k$ is modeled as the product of a fixed static scattering matrix $S_0$ and a time-varying diagonal rotation matrix $D_k(P)$ . This design simplifies hardware implementation while maintaining the unitary property.
Signal Model: The received signal over multiple frames and blocks is formulated as a fourth-order PARATUCK-(2,4) tensor. This model captures the interaction between the UT-RIS channel, the RIS-BS channel, the scattering matrix, and the data symbols.

B. Proposed Receivers

The paper derives two distinct semi-blind receivers based on different tensor reshaping strategies:

PAKRON Receiver (Two-Stage Approach):
- Concept: Transforms the fourth-order PARATUCK model into a third-order PARAFAC model.
- Stage I: Uses a Bilinear Alternating Least Squares (BALS) algorithm to jointly estimate the data symbol matrix and a structured matrix $\Omega$ (which is the Kronecker product of the symbol matrix and the equivalent channel).
- Stage II: Performs a Kronecker factorization on the estimated $\Omega$ to decouple and recover the individual channel matrices (UT-RIS and RIS-BS) and the data symbols.
- Trade-off: Lower computational complexity per iteration but prone to error propagation from Stage I to Stage II.
TUCKER Receiver (Single-Stage Approach):
- Concept: Preserves the original fourth-order structure of the received signal, reformulating it as a fourth-order TUCKER decomposition.
- Algorithm: Uses a Trilinear Alternating Least Squares (TALS) algorithm to iteratively estimate the channel matrices and symbols simultaneously in a single stage.
- Trade-off: Higher computational complexity but avoids error propagation, leading to better accuracy, especially in low SNR regimes.

3. Key Contributions

Novel Semi-Blind Framework: First to propose a semi-blind joint channel and symbol estimation scheme specifically for BD-RIS that eliminates the need for pilot sequences by exploiting data symbols directly.
Dual Receiver Algorithms:
- Development of the PAKRON receiver (PARAFAC-based, two-stage).
- Development of the TUCKER receiver (TUCKER-based, single-stage).
Identifiability Analysis: Derivation of necessary and sufficient conditions (Kruskal rank conditions) for the unique recovery of channels and symbols, providing guidelines for system parameter selection (e.g., number of data blocks $K$ , RIS elements $N$ ).
Complexity vs. Performance Trade-off: A comprehensive analysis showing that while the TUCKER receiver is computationally heavier, it offers superior performance by preserving the full tensor diversity and avoiding error propagation.

4. Simulation Results

Extensive numerical experiments were conducted to evaluate the proposed methods against benchmarks (Pilot-assisted LS, BTKF, BTALS).

Channel Estimation Accuracy (NMSE):
- The TUCKER receiver consistently outperforms the PAKRON receiver, particularly in low-to-moderate SNR regimes, due to the absence of error propagation.
- Both proposed semi-blind methods significantly outperform traditional Least Squares (LS) methods.
- The PAKRON receiver shows sensitivity to the number of RIS groups ( $Q$ ) at low SNR, whereas the TUCKER receiver is more robust.
Symbol Error Rate (SER):
- The TUCKER receiver achieves SER performance close to the ideal "clairvoyant" Zero-Forcing receiver (which assumes perfect channel knowledge).
- The PAKRON receiver suffers performance degradation in SER due to the two-stage error propagation.
Spectral Efficiency: Both proposed methods achieve higher spectral efficiency than pilot-assisted methods because they do not require a dedicated training phase.
Computational Complexity:
- PAKRON requires fewer floating-point operations (FLOPs) per iteration and converges faster in runtime, despite requiring more iterations.
- TUCKER is computationally more complex due to the inversion of larger matrices in the single-stage process but provides higher accuracy.

5. Significance

This work is significant for the advancement of 6G wireless communications and BD-RIS technology:

Overcoming Pilot Overhead: By enabling semi-blind estimation, the method drastically reduces the training overhead required for complex BD-RIS systems, which is critical for high-mobility scenarios and massive MIMO.
Handling Mobility: The framework explicitly accounts for time-varying UT-RIS channels, making it suitable for dynamic environments where traditional static channel models fail.
Theoretical Foundation: The derivation of identifiability conditions and the comparison of PARATUCK vs. TUCKER decompositions provide a solid theoretical basis for future receiver design in reconfigurable surface-assisted networks.
Practical Implementation: The proposed scattering matrix design (static + diagonal rotation) offers a practical path for hardware implementation, balancing performance gains with hardware complexity.

In conclusion, the paper demonstrates that tensor-based semi-blind receivers are a viable and superior alternative to pilot-assisted methods for BD-RIS, offering a flexible trade-off between computational complexity and estimation accuracy depending on the specific system constraints.