⚡ electrical engineering

Multiuser OTFS Channel Parameter Estimation Toward Grid-Independent Regime

This paper proposes a multi-user pilot cyclic prefix design for OTFS systems that enables high-resolution, grid-independent channel parameter estimation through extended W-MUSIC and matrix pencil methods, demonstrating a tradeoff where W-MUSIC excels at low SNR while the matrix pencil approach offers superior accuracy and lower complexity at moderate-to-high SNR.

Original authors: Hanning Wang, Rong-Rong Chen, Arman Farhang

Published 2026-05-07

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Hanning Wang, Rong-Rong Chen, Arman Farhang

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

Imagine you are at a crowded party where everyone is trying to talk to you at once. In a normal room, their voices would blend into a confusing roar. But in this specific scenario, imagine that every person has a unique "voice pitch" (like a specific musical note) that they use to speak. Even though they are all talking at the same time, you can separate them because their voices occupy different parts of the musical scale.

This paper is about solving a similar problem in wireless communication, specifically for a new technology called OTFS (Orthogonal Time Frequency Space). OTFS is designed to work well when things are moving fast (like high-speed trains or drones), where the signal gets scrambled by speed and distance.

Here is the breakdown of what the researchers did, using simple analogies:

1. The Problem: The "Messy" Signal

In high-speed wireless communication, signals bounce off buildings and move with the user. This creates two types of confusion:

Delay: The signal takes different amounts of time to arrive (like an echo).
Doppler: The signal changes pitch because of speed (like a siren passing by).

When multiple users (like four different phones) try to send data at the same time, their signals mix together. Traditional methods try to guess the signal by looking at a fixed "grid" (like a ruler with fixed markings). But real-world signals often fall between the markings (like a ruler that only has inch marks, but the object is 1.5 inches long). This causes errors.

2. The Solution: The "Special Party Invitation" (MU-PCP)

The authors introduced a clever way to organize the users, called MU-PCP (Multi-User Pilot Cyclic Prefix).

The Analogy: Imagine giving each of the four users a different "seat" in the Doppler dimension (the pitch dimension).
How it works: They assign each user a specific offset, like a unique musical note. Even though they are all sending data, the system knows that User A's signal is always shifted by "Note 1," User B by "Note 2," and so on.
The Result: This keeps the signals organized enough that the receiver can still hear the "pure" structure of the message, even with multiple people talking.

3. The Two New Tools: "The Smart Ear" and "The Magic Wand"

Once the signals are organized, the researchers needed two new tools to figure out exactly where the signals came from (delay) and how fast they were moving (Doppler). They developed two methods:

Method A: MU-W-MUSIC (The "Smart Ear")

How it works: This method is like a very sensitive ear that listens to the "noise" in the room to figure out where the voices are coming from. It uses a technique called "Weighted MUSIC."
The Catch: It's very good at listening when the room is loud and noisy (Low Signal-to-Noise Ratio). However, it still relies on a "grid" (a ruler) to some extent, so it's not perfectly precise if the signal falls exactly between the grid lines.
Best for: Noisy environments where you need a reliable guess.

Method B: MU-MP (The "Magic Wand")

How it works: This method uses something called a "Matrix Pencil." Imagine holding a pencil and spinning it; it creates a perfect circle. This method treats the signal like a mathematical shape that can be spun and analyzed to find the exact location without needing a ruler or a grid at all.
The Benefit: It is grid-independent. It doesn't care if the signal falls between the lines; it calculates the exact number.
The Trade-off: It is much faster to compute (less brain power needed) and more accurate when the signal is clear (High Signal-to-Noise Ratio). However, if the room is extremely noisy, it might get confused.

4. The Results: Speed vs. Accuracy

The researchers tested both tools in a simulation with four users. Here is what they found:

In a Noisy Room (Low SNR): The "Smart Ear" (MU-W-MUSIC) was better. It could still hear the signals clearly when the "party" was chaotic.
In a Quiet Room (High SNR): The "Magic Wand" (MU-MP) was the winner. It found the exact location of the signals with much higher precision.
The Speed: The "Magic Wand" was about 10 times faster (computationally cheaper) than the "Smart Ear." It required significantly less processing power to get a great result.

Summary

The paper presents a new way to organize wireless signals from multiple users so they don't crash into each other. They then built two different "detectors" to find the exact speed and distance of these signals:

One detector is robust (good in bad conditions) but a bit slower and less precise.
The other detector is fast and ultra-precise (great in good conditions) and doesn't need a pre-made grid to work.

The authors conclude that by using these tools, we can get a much clearer picture of the wireless channel, which is essential for reliable communication in high-speed scenarios.

Technical Summary: Multiuser OTFS Channel Parameter Estimation Toward Grid-Independent Regime

Problem Statement
Accurate channel estimation in high-mobility wireless environments is critical, as channels become doubly selective in the time-frequency (TF) domain, rendering conventional TF-domain methods ineffective. While Orthogonal Time Frequency Space (OTFS) offers a structured representation in the delay-Doppler (DD) domain suitable for parametric estimation, existing methods face significant challenges in multiuser settings. Current approaches are broadly categorized into grid-based and grid-independent techniques. Grid-based methods suffer from basis mismatch when dealing with fractional delay and Doppler shifts. Conversely, grid-independent methods like Weighted MUSIC (W-MUSIC) and Matrix Pencil (MP) offer high resolution but have not been effectively extended to multiuser OTFS due to inter-user interference and the lack of a signal structure that preserves separability. Furthermore, a tradeoff exists between the robustness of subspace methods at low Signal-to-Noise Ratio (SNR) and the high-resolution accuracy of parametric methods at moderate-to-high SNR.

Methodology
The authors propose a framework for multiuser OTFS channel parameter estimation that leverages a Multi-User Pilot with Cyclic Prefix (MU-PCP) design. This design multiplexes users along the Doppler dimension while preserving a separable exponential structure in the received signal.

System Model: The paper adopts the MU-PCP structure where each user is assigned a specific Doppler offset ( $k_q$ ). This allows the received pilot signal in the TF domain to be modeled as a sum of separable complex exponentials, where delay and Doppler shifts appear as distinct steering vectors. This structure enables user separation and reformulates the estimation problem as a joint parametric problem over fractional delays and Dopplers.
MU-W-MUSIC: The authors extend the W-MUSIC algorithm to the multiuser uplink. This method constructs a noise subspace via spatial smoothing and eigenvalue decomposition. It approximates the Doppler null-spectrum polynomial using a truncated Fourier series (weighted least squares) to find roots near the unit circle. While it reduces computational load compared to standard root-MUSIC, it retains a "mild" grid dependency as polynomial roots are evaluated over discretized samples.
MU-MP (Matrix Pencil): To achieve a fully grid-independent solution, the authors develop a Matrix Pencil-based method. This approach constructs a structured block Hankel matrix from the received signal. By exploiting the shift-invariance property of the Doppler dimension, it formulates a generalized eigenvalue problem to directly estimate Doppler poles. Subsequently, delay parameters are estimated by projecting the received signal onto the estimated Doppler subspace and analyzing phase differences.
Channel Gain Recovery: Once fractional delay and Doppler parameters are estimated for all paths, channel gains are recovered using a Least Squares (LS) formulation based on a parametric dictionary constructed from the estimated parameters.

Key Contributions

MU-PCP Pilot Design: Introduction of a pilot design that preserves a separable exponential structure with user-specific Doppler offsets, enabling high-resolution parametric estimation in multiuser scenarios.
Extension of W-MUSIC: Adaptation of W-MUSIC to multiuser OTFS, providing a computationally efficient approach with mild grid dependency.
Grid-Independent MP Estimator: Development of a fully grid-independent Matrix Pencil method using a structured block Hankel construction and reduced pencil formulation. This method achieves direct estimation of DD parameters with improved accuracy at moderate-to-high SNR.
Robustness-Complexity Tradeoff Analysis: Demonstration that while MU-W-MUSIC offers superior robustness at low SNR, the MU-MP method achieves higher estimation accuracy at moderate-to-high SNR with significantly lower computational complexity.

Results
Numerical simulations were conducted on a multiuser OTFS system with 4 users, considering both integer and fractional delay/Doppler shifts.

Estimation Accuracy: In the low-SNR regime (0–5 dB), MU-W-MUSIC outperforms MU-MP in both Doppler and delay estimation. However, as SNR increases beyond 10–15 dB, MU-MP demonstrates superior performance. Specifically, for delay estimation, MU-MP achieves a 5 dB SNR gain over MU-W-MUSIC at an RMSE of $10^{-3}$ .
Channel Reconstruction: Both methods achieve full channel matrix reconstruction with RMSE below $10^{-3}$ when SNR exceeds 20 dB. MU-MP provides an additional ~3 dB gain at RMSE levels around $10^{-4}$ .
Computational Complexity: The MU-MP method requires approximately one order of magnitude fewer complex multiplications (CMs) than MU-W-MUSIC ( $2.68 \times 10^7$ vs. $2.65 \times 10^8$ ). This reduction is primarily due to the absence of the spectral sampling stage in MU-MP, which dominates the complexity of MU-W-MUSIC.

Significance
The paper claims that by leveraging the MU-PCP structure, it successfully bridges the gap between structured parametric estimation and multiuser OTFS systems. The proposed framework enables high-resolution estimation of fractional delay and Doppler parameters without relying on discretized grids. The work highlights a practical tradeoff: MU-W-MUSIC is suitable for low-SNR environments due to its robustness, while MU-MP is preferred for moderate-to-high SNR scenarios where high accuracy and low computational complexity are paramount. This dual-method approach provides a flexible foundation for integrated sensing and communication in future high-mobility wireless systems.