Diversity Analysis of Symbol-by-Symbol Linear Equalizers

Ali Tajer, Aria Nosratinia, Naofal Al-Dhahir

TL;DR

This paper provides a rigorous theoretical proof that symbol-by-symbol Minimum Mean-Square Error (MMSE) linear equalizers achieve the full frequency diversity of a multipath channel, matching the diversity performance of the much more complex Maximum Likelihood (ML) receivers. In contrast, the authors demonstrate that Zero-Forcing (ZF) equalizers are fundamentally limited, always resulting in a diversity order of one regardless of the channel's complexity.

What Problem Does This Paper Address?

In broadband wireless communication, signals travel through multiple paths, leading to frequency-selective fading and Inter-Symbol Interference (ISI). While ISI is usually a hurdle, it also provides "frequency diversity"—essentially multiple "looks" at the same signal—which can be exploited to make the connection more reliable.

The optimal way to handle this is Maximum Likelihood Sequence Detection (MLSD), but its computational complexity grows exponentially with the channel's memory and spectral efficiency. Engineers prefer simple linear equalizers (like ZF and MMSE) because they are computationally "cheap." However, for decades, it wasn't fully understood if these simpler receivers were "throwing away" the channel's inherent diversity. This paper aims to mathematically characterize the diversity order achieved by these common linear equalizers to see if they can truly compete with optimal receivers.

Methods

The authors employ a mix of high-level communication theory and rigorous mathematical analysis:

  • System Modeling: They use a quasi-static ISI channel model with memory length $\nu$. They represent the system in the $D$-domain (transform domain) to analyze the signal-to-noise ratio (SNR) at the equalizer's output.
  • Outage Probability Analysis: Diversity order is defined by how fast the error probability (or outage probability) drops as the SNR increases. The authors analyze the "exponential order" of these probabilities.
  • Hardy’s Inequality: To prove the MMSE results, the authors utilize a sophisticated tool from mathematical analysis—Hardy’s inequality—to find a lower bound on the $L_1$ norm of exponential sums. This allows them to bound the performance of the MMSE equalizer.
  • Induction & Monotone Convergence: They first prove full diversity for low data rates (spectral efficiencies) and then use the Monotone Convergence Theorem to extend the proof to all possible data rates.
  • ZF Analysis: For the Zero-Forcing case, they use the properties of Gaussian random variables to show that the noise enhancement inherent in ZF equalization effectively "kills" any diversity gain beyond the first order.

Datasets

As a theoretical paper, it does not use real-world datasets. Instead, the authors validate their mathematical proofs using Monte Carlo simulations of synthetic Rayleigh fading channels:

  • Channel Configurations: Two-tap ($\nu=1$) and three-tap ($\nu=2$) ISI channels.
  • Evaluation Metric: Average Bit Error Probability (BER) vs. SNR (dB).
  • Parameters: Various spectral efficiencies ($R = 1, 2, 3, 4$ bits/sec/Hz) were tested to show that diversity order is independent of the data rate.

Key Insights and Ideas

  • The "Balance" of MMSE: The core "aha" moment is that MMSE equalizers succeed because they strike a balance between reducing ISI and limiting noise enhancement. This balance is sufficient to capture all available frequency diversity ($\nu+1$).
  • The Failure of ZF: Zero-Forcing equalizers fail because they focus only on removing ISI. In doing so, they can massively amplify noise when the channel has "notches" (frequencies with very low gain). This noise enhancement is so severe that it limits the diversity order to 1, no matter how many paths the channel has.
  • Rate Independence: A surprising finding is that the diversity order for these equalizers is a fundamental property of the equalizer type and the channel memory; it does not change even if you increase the transmission rate (spectral efficiency).
  • MMSE vs. MLSD: While MMSE achieves the same diversity order (slope of the error curve) as the optimal MLSD, it still suffers from a "coding gain" loss (the curve is shifted to the right). It's just as robust, but slightly less efficient.

Key Results

  • MMSE Diversity Order: Proved to be $\nu + 1$. If a channel has 3 taps, the error rate drops as $SNR^{-3}$.
  • ZF Diversity Order: Proved to be exactly 1. Regardless of the number of taps, the error rate only drops as $SNR^{-1}$.
  • Simulation Match: The simulation results (Figures 1 and 2 in the paper) perfectly align with the theory. In the MMSE plots, the curves for $\nu=2$ are significantly steeper than those for $\nu=1$. In the ZF plots, all curves have the same shallow slope regardless of $\nu$.

Limitations and Future Work

  • Infinite Length Assumption: The proofs assume "infinite-length" equalizers (ideal filters). In practice, equalizers use Finite Impulse Response (FIR) filters. While the authors suggest the results should hold, a formal analysis of finite-length constraints would be a valuable next step.
  • Decision Feedback Equalizers (DFE): This paper focuses on linear equalizers. The authors mention that symbol-by-symbol decision feedback equalizers are also widely used, and extending this diversity analysis to DFEs is an open area for research.
  • Non-Gaussian Noise: The analysis assumes additive white Gaussian noise (AWGN). Exploring how these equalizers perform in the presence of non-Gaussian interference or impulsive noise could provide more practical insights for cellular environments.

Summary updated

Abstract

In frequency-selective channels linear receivers enjoy significantly-reduced complexity compared with maximum likelihood receivers at the cost of performance degradation which can be in the form of a loss of the inherent frequency diversity order or reduced coding gain. This paper demonstrates that the minimum mean-square error symbol-by-symbol linear equalizer incurs no diversity loss compared to the maximum likelihood receivers. In particular, for a channel with memory ν, it achieves the full diversity order of (ν+1) while the zero-forcing symbol-by-symbol linear equalizer always achieves a diversity order of one.
BibTeX
@article{tajer2011diversity,
  author = {Ali Tajer and Aria Nosratinia and Naofal Al-Dhahir},
  title = {{Diversity Analysis of Symbol-by-Symbol Linear Equalizers}},
  journal = {IEEE Trans. Communications},
  year = {2011},
  url = {https://openreview.net/forum?id=kSZwkB-u4Nd},
}

Loading related content...

Figure