Collision Resolution in RFID Systems Using Antenna Arrays and Mixed Source Separation

Basic Information

• Paper ID: 2511.15490
• Title: Collision Resolution in RFID Systems Using Antenna Arrays and Mixed Source Separation
• Authors: Mohamed Siala (MEDIATRON Lab., SUP'COM, University of Carthage, Tunisia), Noura Sellami (LETI Lab., ENIS, University of Sfax, Tunisia)
• Classification: eess.SP (Signal Processing), cs.IT (Information Theory), math.IT (Information Theory)
• Submission Date: November 19, 2025 to arXiv
• Paper Link: https://arxiv.org/abs/2511.15490

Abstract

This paper proposes an efficient mixed source separation (MSS) algorithm to resolve tag collision problems in RFID reader systems equipped with antenna arrays. The authors first introduce a method utilizing the Zero-Crossing Modulus (ZCM) criterion through gradient descent to separate colliding tags without requiring pilot symbols. The research demonstrates that designing objective functions using the ZCM property alone leads to significant ambiguity when determining beamformers for tag message recovery. To address this limitation, the authors propose a more sophisticated approach based on a hybrid objective function that introduces new disambiguation criteria beyond the ZCM criterion.

Research Background and Motivation

1. Core Problem to Be Solved

Tag collision problem in RFID systems: When multiple RFID tags are activated simultaneously, their messages collide and interfere with each other at the reader, necessitating retransmission of tag identity information, resulting in bandwidth waste and increased identification delays.

2. Problem Significance

• RFID technology plays a critical role in supply chain management, security, access control, and transportation
• Tag collision is a core bottleneck constraining RFID system performance
• Efficient collision resolution solutions are essential for improving system throughput and reducing identification latency

3. Limitations of Existing Methods

Existing collision resolution methods fall into two main categories:

• Network layer methods: Such as tree-splitting algorithms or ALOHA protocols, addressing problems through collision avoidance techniques
• Signal processing methods: Such as source separation techniques

Particularly, the Algebraic ZCM Algorithm (AZCMA) proposed in reference 6 has critical defects:

• Requires solving matrix pencil problems with high computational complexity
• Produces severe ambiguity under perfect or near-perfect frequency synchronization, leading to spurious beamformers
• Depends on system imperfections such as oscillator drift and phase fluctuations

4. Research Motivation

• Most RFID tags lack training symbols, making channel estimation difficult
• Mixed source separation (MSS) is a viable approach to overcome tag identification problems
• There is a need to develop algorithms that work effectively under perfect synchronization conditions

Core Contributions

1. Revealed the inherent ambiguity problem of the ZCM criterion: Systematically analyzed and proved that using the ZCM criterion alone leads to spurious beamformers when tag signals are frequency-synchronized with the receiver
2. Proposed new disambiguation criteria: Introduced a new criterion based on the product of three consecutive symbols $\pi[n] = \hat{s}[n-1]\hat{s}^*[n]\hat{s}[n+1]$, which is identically zero for true beamformers but takes non-zero values with non-zero probability for spurious beamformers
3. Designed hybrid objective function: Combined the ZCM criterion and new disambiguation criterion into a weighted hybrid objective function $J_{01}(w) = \lambda J_0(w) + (1-\lambda)J_1(w)$
4. Developed gradient descent-based optimization algorithm: Proposed complete gradient computation formulas and iterative update strategies, avoiding the need to solve matrix pencil problems
5. Verified algorithm robustness: Through simulation experiments, demonstrated the effectiveness of the proposed method under perfect synchronization and scenarios with frequency offset/random phase variations

Detailed Methodology

Task Definition

Input:

• $M \times N$ received data matrix $\mathbf{X}$, where $M$ is the number of receive antennas and $N$ is the number of symbols in a packet
• $\mathbf{X} = \mathbf{AS} + \mathbf{N}$, containing superimposed signals from $d$ tags

Output:

• $d$ beamforming vectors $\mathbf{w}_i$ of size $M \times 1$ ($1 \leq i \leq d$)
• Recovered tag signals $\hat{\mathbf{s}}_i = \mathbf{w}_i^H\mathbf{X}$

Constraints:

• Channel matrix $\mathbf{A}$ has full column rank (overdetermined or exactly determined system)
• Tag signals satisfy ZCM property: $|s_i[n]| \in \{0, 1\}$
• Manchester encoding is employed

System Model

Signal Model

The $n$-th symbol of the $i$-th tag is modeled as: $s_i[n] = b_i[n] \exp(j\phi_i[n]), \quad 1 \leq n \leq N$

where $b_i[n] \in \{0, 1\}$ is the binary message and $\phi_i[n]$ is a random phase.

Manchester Encoding Rules:

• Logical 1: $(b_i[2k], b_i[2k+1]) = (1, 0)$
• Logical 0: $(b_i[2k], b_i[2k+1]) = (0, 1)$

Narrowband Channel Model

The received signal matrix is expressed as: $\mathbf{X} = \mathbf{AS} + \mathbf{N}$

where:

• $\mathbf{A} = [\mathbf{a}_1, \mathbf{a}_2, \cdots, \mathbf{a}_d]$: $M \times d$ channel matrix
• $\mathbf{S}$: $d \times N$ signal matrix
• $\mathbf{N}$: $M \times N$ noise matrix

Ambiguity Analysis

Mathematical Expression of ZCM Criterion

The ZCM property can be compactly expressed as: $s[n](|s[n]|^2 - 1) = 0$

Under ideal conditions (no noise, perfect beamforming, constant phase $\phi_i[n] = \phi_i$), the beamformer output is: $\hat{s}_i[n] = \mathbf{w}_i^H\mathbf{x}[n] = e^{j\theta_i}s_i[n] = e^{j\psi_i}b_i[n]$

where $\psi_i = \theta_i + \phi_i$.

Construction of Spurious Beamformers

Two-tag case: For any two tag beamformers $\mathbf{w}_{i_1}$ and $\mathbf{w}_{i_2}$, construct spurious beamformers: $\mathbf{w}^{\pm} = e^{j\psi_{i_1}}\mathbf{w}_{i_1} + e^{j(\pm 2\pi/3 + \psi_{i_2})}\mathbf{w}_{i_2}$

The corresponding output is: $(\mathbf{w}^{\pm})^H\mathbf{x}[n] = b_{i_1}[n] + e^{\mp j2\pi/3}b_{i_2}[n]$

Key Finding: As shown in Table I, this spurious beamformer output still satisfies the ZCM property (magnitude is always 0 or 1), and thus may become a solution to AZCMA.

Three-tag case: Similarly, one can construct: $\mathbf{w}^{\pm} = e^{j\psi_{i_1}}\mathbf{w}_{i_1} + e^{j(\pm 2\pi/3 + \psi_{i_2})}\mathbf{w}_{i_2} + e^{j(\mp 2\pi/3 + \psi_{i_3})}\mathbf{w}_{i_3}$

The output still satisfies the ZCM property, leading to additional ambiguities.

New Disambiguation Criterion

Define the product of three consecutive symbols: $\pi[n] = \hat{s}[n-1]\hat{s}^*[n]\hat{s}[n+1] = \mathbf{w}^H\mathbf{x}[n-1]\mathbf{x}^H[n]\mathbf{w}\mathbf{w}^H\mathbf{x}[n+1]$

Key Properties:

• For true beamformers: $\pi[n] \equiv 0$ (guaranteed by the alternating property of Manchester encoding)
• For spurious beamformers: $\pi[n]$ takes non-zero values with non-zero probability

Table II demonstrates that when using spurious beamformer $\mathbf{w}^+$, multiple combinations among 16 cases result in $\pi[n] \neq 0$.

Hybrid Objective Function Design

ZCM Objective Function

$J_0(\mathbf{w}) = \frac{1}{N}\sum_{n=1}^{N}|\hat{s}[n]|^2(|\hat{s}[n]|^2 - 1)^2$

The gradient is: $\nabla J_0(\mathbf{w}) = \frac{1}{N}\sum_{n=1}^{N}c[n]^*\mathbf{x}[n]$

where: $c[n] = 2(|\mathbf{w}^H\mathbf{x}[n]|^2 - 1)(2|\mathbf{w}^H\mathbf{x}[n]|^2 + 1)(\mathbf{w}^H\mathbf{x}[n])$

New Disambiguation Objective Function

$J_1(\mathbf{w}) = \frac{1}{N-2}\sum_{n=2}^{N-1}|\hat{s}[n-1]\hat{s}[n]^*\hat{s}[n+1]|^2$

The gradient is: $\nabla J_1(\mathbf{w}) = \frac{1}{N-2}\sum_{n=2}^{N-1}(c_-[n]^*\mathbf{x}[n-1] + c_0[n]^*\mathbf{x}[n] + c_+[n]^*\mathbf{x}[n+1])$

where: $c_-[n] = 2|\mathbf{w}^H\mathbf{x}[n]|^2|\mathbf{w}^H\mathbf{x}[n+1]|^2(\mathbf{w}^H\mathbf{x}[n-1])$$c_0[n] = 2|\mathbf{w}^H\mathbf{x}[n-1]|^2|\mathbf{w}^H\mathbf{x}[n+1]|^2(\mathbf{w}^H\mathbf{x}[n])$$c_+[n] = 2|\mathbf{w}^H\mathbf{x}[n-1]|^2|\mathbf{w}^H\mathbf{x}[n]|^2(\mathbf{w}^H\mathbf{x}[n+1])$

Hybrid Objective Function

$J_{01}(\mathbf{w}) = \lambda J_0(\mathbf{w}) + (1-\lambda)J_1(\mathbf{w})$

where $\lambda \in (0, 1)$ is a weighting parameter used to balance the two criteria.

Gradient Descent Algorithm

The iterative update formula is: $\mathbf{w}[l+1] = \mathbf{w}[l] - \mu\nabla J_{01}(\mathbf{w})|_{\mathbf{w}=\mathbf{w}[l]}$

where:

• $\mu$: step size
• $l$: iteration number
• $\nabla J_{01}(\mathbf{w}) = \lambda\nabla J_0(\mathbf{w}) + (1-\lambda)\nabla J_1(\mathbf{w})$

Experimental Setup

Simulation Parameters

• Packet size: $N = 100$ symbols
• Signal-to-noise ratio: SNR = $E_b/N_0 = 20$ dB
• Step size: $\mu = 10^{-2}$
• Number of iterations: $L \in \{100, 200, 400, 800\}$
• Success criterion: Tag message is correctly identified

Experimental Scenarios

1. Perfect frequency synchronization: $\phi_i[n] = \phi_i$ (constant phase)
2. Frequency offset: $s_i[n] = b_i[n]\exp(2j\pi\delta F_in)$, where $\delta F_i \sim U[-\delta F/2, \delta F/2]$
3. Random phase: $\phi_i[n] \sim U[0, 2\pi]$

System Configuration

• Number of tags: $d \in \{2, 3, 4\}$
• Number of receive antennas: $M \in \{2, 3, 4, 5\}$
• Considers both overdetermined ($M > d$) and exactly determined ($M = d$) systems

Experimental Results

Main Results Analysis

1. Impact of Weight Parameter $\lambda$ (Figure 1a)

Configuration: $d=2$, $M=2$, perfect synchronization

Key Findings:

• $\lambda = 1$ (pure ZCM criterion): Success rate extremely low (<10%), validating the failure of reference 6 method
• $\lambda > 0.6$: Success rate still very low, ambiguity problem severe
• $0.025 < \lambda < 0.6$: Success rate significantly improved
• $\lambda = 0$ (pure new criterion): Success rate zero, indicating the new criterion requires combination with ZCM criterion
• Optimal range: As iteration count $L$ increases, the $\lambda$ range with high success rate expands; at $L=800$, success rate approaches 100% in range $0.05 < \lambda < 0.5$

Conclusion: The necessity of hybrid strategy is verified; system shows good robustness to $\lambda$ selection.

2. Performance Under Different System Configurations (Figure 1b)

Configuration: $L=800$, perfect synchronization

Results:

• Exactly determined systems ($M=d$):
  • $(d=2, M=2)$: Success rate >90% when $0.025 < \lambda < 0.4$
  • $(d=3, M=3)$: Good performance when $0.05 < \lambda < 0.3$
  • $(d=4, M=4)$: Effective when $0.1 < \lambda < 0.25$
• Overdetermined systems ($M>d$):
  • $(d=2, M=3)$, $(d=3, M=4)$, $(d=4, M=5)$: Optimal $\lambda$ range shifts toward smaller values, with better performance at $\lambda < 0.2$
  • Overdetermined systems provide more spatial degrees of freedom, enhancing separation capability

Conclusion: Algorithm shows good adaptability to different system configurations; overdetermined systems perform better.

3. Impact of Frequency Offset and Random Phase (Figure 1c)

Configuration: $d=2$, $M=2$, $L=800$

Results:

• $\delta F = 0$ (perfect synchronization): Reference 6 method ($\lambda=1$) fails; proposed method achieves success rate >95% when $0.05 < \lambda < 0.4$
• $\delta F = 0.005$: Performance slightly improved
• $\delta F \geq 0.01$: As frequency offset increases, performance at high $\lambda$ values improves, as frequency offset helps ZCM criterion eliminate ambiguity
• Random phase: Performance optimal for all $\lambda$ values (success rate approaches 100%), as phase randomness naturally eliminates ambiguity

Key Insights:

1. The proposed method remains effective under perfect synchronization (most challenging scenario), while reference 6 method completely fails
2. The proposed method performs better with system imperfections (frequency offset, phase fluctuations)
3. Algorithm is robust to system imperfections, not dependent on artificially introduced imperfections

Performance Comparison Summary

Performance Improvement: Under perfect synchronization conditions, success rate improved from <10% to >95% (improvement exceeding 85 percentage points).

Related Work

1. RFID Collision Avoidance Techniques

• Tree-splitting algorithms: Avoid collisions through time division
• ALOHA protocol and variants 3: Random access strategies
• Limitations: Increased identification delay, low spectral efficiency

2. Signal Processing Methods

• Blind source separation 5: Separation using signal statistical properties
• Independent Component Analysis (ICA) 7: Assumes statistical independence of source signals
• Underdetermined blind separation 8: Handles cases where source count exceeds sensor count

3. ZCM-based Methods

• Mindikoglu and van der Veen (2008) 6:
  • First application of ZCM criterion to RFID collision resolution
  • Uses algebraic method to solve matrix pencil problems
  • This paper identifies its inherent defects: Fails under perfect synchronization

4. Multi-antenna RFID Systems

• ALOHA combined with MSS 7: Hybrid network and physical layer methods
• Advantages: Spatial diversity gain, improved separation performance

Positioning of This Work

• Inheritance: Utilizes ZCM property for unsupervised separation
• Innovation: Reveals ambiguity problem and proposes systematic solution
• Advantages:
  1. Not dependent on frequency/phase imperfections
  2. Gradient descent avoids matrix pencil solving
  3. Hybrid criterion provides robustness
  4. Combines theoretical analysis with experimental verification

Conclusions and Discussion

Main Conclusions

1. Theoretical Contribution: Systematically revealed the ambiguity problem of ZCM criterion under perfect synchronization conditions, proving that any two or three true beamformers can construct two additional spurious solutions
2. Method Contribution: Proposed new disambiguation criterion based on product of three consecutive symbols, which is identically zero for true solutions and takes non-zero values with non-zero probability for spurious solutions
3. Algorithm Contribution: Designed hybrid objective function and corresponding gradient descent algorithm, avoiding complex matrix pencil solving
4. Experimental Verification:
   • Under perfect synchronization, reference 6 method success rate <10%, proposed method >95%
   • Algorithm shows good robustness to $\lambda$ selection with wide effective range
   • Effective for different system configurations (exactly determined/overdetermined)
   • Better performance with system imperfections

Limitations

1. Computational Complexity: Gradient descent requires multiple iterations (L=800 in experiments), potentially more time-consuming than one-shot algebraic solving
2. Initialization Dependency: Gradient descent performance may depend on initialization strategy, not thoroughly discussed in the paper
3. Incomplete Theoretical Analysis:
   • No convergence proof provided
   • No theoretical analysis of optimal $\lambda$ selection criterion
   • No theoretical guidance for $\lambda$ under different $(d, M)$ configurations
4. Experimental Limitations:
   • Only narrowband channel model considered
   • Large-scale systems not tested ($d, M > 5$)
   • Channel estimation error impact not considered
5. Manchester Encoding Specific: Algorithm design based on alternating property of Manchester encoding; applicability to FM0 encoding, though claimed equivalent, not thoroughly verified
6. Noise Sensitivity: Performance under low SNR conditions not thoroughly explored (only SNR=20dB tested)

Future Directions

While not explicitly stated in the paper, inferrable research directions include:

1. Theoretical Refinement:
   • Convergence analysis and convergence rate study
   • Adaptive selection strategy for optimal weight $\lambda$
   • Unified theoretical framework for different encoding schemes
2. Algorithm Improvement:
   • Fast convergence optimization algorithms (e.g., conjugate gradient, Newton's method)
   • Adaptive step size strategies
   • Joint optimization of multiple beamformers
3. Practical Applications:
   • Joint design with network layer protocols
   • Real-time hardware implementation
   • Scalability for large-scale RFID systems
4. Extended Scenarios:
   • Wideband channel model
   • Mobile tag scenarios
   • Underdetermined systems ($d > M$)

In-Depth Evaluation

Strengths

1. Theoretical Innovation (★★★★★)

• Deep Insight: First systematic revelation of ZCM criterion's ambiguity problem, with constructive proof demonstrating existence of spurious beamformers
• Mathematical Rigor: Tables I and II clearly show how spurious solutions satisfy ZCM property
• Clear Problem Definition: Attributes ambiguity problem to perfect synchronization, grasping the problem essence

2. Clever Method Design (★★★★☆)

• New Criterion Design: Leverages Manchester encoding's alternating property; $\pi[n]$ criterion is simple yet effective
• Hybrid Strategy: Balances advantages of two criteria, avoiding defects of single criterion
• Complete Gradient Derivation: Provides detailed gradient computation formulas, facilitating implementation

3. Sufficient Experiments (★★★★☆)

• Multi-dimensional Verification:
  • Impact of different iteration counts
  • Performance under different system configurations
  • Robustness under different synchronization conditions
• Clear Comparison: Direct comparison with reference 6 highlights improvements
• Comprehensive Parameter Analysis: $\lambda$ selection range thoroughly explored

4. Practical Value (★★★★☆)

• Real-world Significance: Addresses common perfect synchronization scenario in practical systems
• No Artificial Imperfections Required: Not dependent on frequency offset or phase fluctuation
• Strong Implementability: Gradient descent algorithm easy to program

5. Writing Quality (★★★★★)

• Clear Logic: Complete problem→analysis→solution→verification flow
• Precise Expression: Standardized mathematical notation, rigorous formula derivation
• Good Readability: Tables and figures effectively support arguments

Weaknesses

1. Insufficient Theoretical Analysis (★★★☆☆)

• Missing Convergence Proof: Convergence of gradient descent algorithm not theoretically guaranteed
• Absent Optimality Analysis: Global optimality of hybrid objective function not proven
• Lack of Theoretical Guidance for Parameter Selection: $\lambda$ selection mainly relies on experiments, lacking theoretical basis

2. Limited Experimental Design (★★★☆☆)

• Single SNR Value: Only 20dB tested, low SNR performance unexplored
• Limited Scale: Maximum test only $d=4, M=5$; large-scale system performance unknown
• Missing Statistical Significance Tests: No mean and variance reported for multiple runs
• Unaddressed Initialization: Beamformer initialization strategy not explained

3. Method Limitations (★★★☆☆)

• Encoding Specific: Strongly dependent on Manchester encoding's alternating property
• Computational Complexity: Requires 800 iterations; real-time performance may be affected
• Local Optimum Risk: Gradient descent may get stuck in local optimum; no solution provided

4. Insufficient Practical Application Considerations (★★★☆☆)

• Channel Estimation: Assumes channel matrix $\mathbf{A}$ is known or estimable, but estimation error impact not discussed
• Reader Signal Cancellation: Assumes perfect reader signal cancellation, but residuals may exist in practice
• Antenna Spacing Requirements: 16cm spacing may be difficult to implement on portable devices

5. Brief Related Work Discussion (★★★☆☆)

• Incomplete Comparison: Not compared with other blind source separation methods (ICA, sparse coding)
• Insufficient Advantage Analysis: Specific advantages over other MSS methods not detailed

Impact Assessment

Contribution to Field (★★★★☆)

• Theoretical Contribution: Reveals inherent defects of ZCM criterion, providing important insights for subsequent research
• Method Contribution: Provides practical solution directly applicable to RFID systems
• Inspirational Value: Hybrid criterion design approach generalizable to other blind separation problems

Practical Value (★★★★☆)

• Direct Application: Applicable to UHF RFID system collision resolution
• Significant Performance Improvement: Success rate improvement >85% in critical scenarios
• Moderate Implementation Difficulty: Gradient descent algorithm easy to implement

Reproducibility (★★★☆☆)

• Strengths:
  • Detailed algorithm description, complete formulas
  • Clear parameter settings
• Weaknesses:
  • No code or pseudocode provided
  • Initialization strategy not specified
  • Stopping criterion not clearly defined

Potential Impact

• Short-term: Likely rapid adoption and verification by RFID research community
• Medium-term: May inspire blind separation research in other communication systems
• Long-term: Hybrid criterion design thinking may become universal paradigm for solving ambiguity problems

Applicable Scenarios

Most Suitable Scenarios (★★★★★)

1. High-quality synchronization systems:
   • RFID tags with high-precision oscillators
   • Short-range communication (small synchronization error)
   • Indoor environments (stable channels)
2. Fixed tag applications:
   • Warehouse management
   • Library systems
   • Asset tracking
3. Overdetermined systems:
   • Reader with more antennas than tags
   • Sufficient space for antenna array deployment

Applicable but Requiring Improvement (★★★☆☆)

1. Mobile tags: Need to consider Doppler effect
2. Large-scale systems: Scalability needs verification
3. Low SNR environments: Need enhanced noise robustness

Inapplicable Scenarios (★★☆☆☆)

1. Underdetermined systems ($d > M$): Requires additional sparsity assumptions
2. Wideband systems: Narrowband assumption invalid
3. Extreme real-time requirements: Iterative algorithm may be too slow
4. Non-Manchester encoding: New criterion's effectiveness needs re-verification

Comprehensive Scoring

Recommended Readers

1. Strongly Recommended: RFID system researchers, blind source separation researchers, antenna array signal processing researchers
2. Recommended: Wireless communication researchers, IoT system designers
3. Optional: Signal processing students (as gradient descent application case study)

Key References

6 A. F. Mindikoglu and A.-J. van der Veen, "Separation of overlapping RFID signals by antenna arrays," ICASSP 2008

• Main comparison target, proposes AZCMA method

5 Z. Luo et al., "A Comprehensive Survey on Blind Source Separation for Wireless Adaptive Processing," IEEE Access 2018

• Blind source separation survey, provides theoretical background for MSS

9 ISO/IEC 18000-6:2013

• RFID standard, defines Manchester encoding and system parameters

Summary Statement

This paper addresses the tag collision problem in RFID systems by proposing a gradient descent algorithm based on hybrid objective functions. The core contribution lies in revealing and resolving the ambiguity problem of existing ZCM criterion under perfect synchronization conditions. By introducing a new disambiguation criterion and combining it with the ZCM criterion, the algorithm achieves >95% success rate in the most challenging scenario, improving over 85 percentage points compared to existing methods.

The paper's main strengths include: (1) Deep theoretical insight, systematically analyzing the mathematical roots of ambiguity; (2) Clever method design with simple yet effective new criterion; (3) Sufficient experimental verification demonstrating algorithm robustness across multiple dimensions.

Main limitations include: (1) Incomplete theoretical analysis, lacking convergence proof and theoretical guidance for parameter selection; (2) Limited experimental scale, untested large-scale systems and low SNR scenarios; (3) High computational complexity potentially affecting real-time applications.

Overall, this is a high-quality research work (8.05/10) providing practical and effective solutions for RFID collision resolution, with important theoretical and practical value for signal processing and wireless communication fields. Subsequent research should focus on theoretical refinement, large-scale system verification, and real-time implementation optimization.