Equalisation with adaptive time lag

Yanmin Gong, C.f.n. CowanIEE 2005 · DOI

TL;DR

This paper tackles the critical but often overlooked problem of selecting the optimal "time lag" (or decision delay) in adaptive equalizers, which significantly impacts performance. The authors introduce a novel adaptive algorithm that dynamically adjusts this time lag to minimize the mean-squared error, leading to improved equalization performance compared to fixed-lag approaches.

What Problem Does This Paper Address?

In digital communications, adaptive equalizers are essential for combating intersymbol interference (ISI) caused by channel distortions. A key parameter in these equalizers is the "time lag" or "decision delay," which essentially determines how the equalizer aligns the received signal with the desired transmitted signal. Choosing the wrong time lag can lead to suboptimal performance, wasting valuable resources. Traditionally, this time lag is either pre-assumed based on channel characteristics or set through experience, neither of which guarantees optimal performance, especially for unknown or time-varying channels. This paper aims to address this by proposing an adaptive approach to find the best time lag automatically.

Methods

The core of the paper lies in two main contributions:

  • New Interpretation of MMSE Equalizer: The authors re-interpret the Minimum Mean-Square Error (MMSE) equalizer. This new perspective allows them to better understand the relationship between the MMSE performance and the time lag, revealing that the MMSE is generally a concave function of the time lag, though local minima can exist.
  • Novel Adaptive Time Lag Algorithm: Based on the insights from the new interpretation, they propose an adaptive algorithm that uses gradient search to find the optimal time lag. A key innovation here is the introduction of a "pseudo fractional time lag." This allows for smoother adaptation by treating the time lag as a continuous variable during the adaptation process, with the true integer time lag being derived from its integer part. This approach helps overcome the discrete nature of the time lag and improves convergence. The algorithm also incorporates a "leaky factor" to prevent the lag from drifting to unnecessarily large values and a "tap-weight shift" to mitigate MSE spikes when the time lag changes.

Datasets

Not discussed in the paper. The simulations use synthetic data generated with specified channel characteristics and signal-to-noise ratios.

Key Insights and Ideas

  • Time Lag is Crucial: The paper strongly emphasizes that the time lag is not a minor tuning parameter but a significant factor that directly impacts the efficiency and performance of adaptive equalizers.
  • "Flat Line" MMSE: When plotting MMSE against time lag, a "flat line" region often appears around the optimal lag. The paper argues that in such cases, it's more desirable to choose the smallest lag within this flat region to minimize delay without sacrificing performance, especially for time-critical systems.
  • Pseudo Fractional Time Lag for Smoothness: Treating the time lag as a continuous (fractional) variable during adaptation, even though the true lag is an integer, is a clever way to achieve smoother and more robust convergence.
  • Tap-Weight Shift for Stability: The inclusion of a tap-weight shift mechanism is crucial for preventing sudden MSE increases when the time lag is updated, ensuring a more stable equalization process.

Key Results

The paper demonstrates through numerical simulations that the proposed adaptive time lag algorithm:

  • Converges to the Optimum Time Lag: The algorithm successfully finds the optimal time lag for various channel types (mixed-phase, minimum-phase, maximum-phase).
  • Achieves Lower MMSE: The adaptive time lag equalizer consistently achieves lower MMSE compared to equalizers with a fixed, suboptimal time lag. For instance, in the maximum-phase channel example, the adaptive equalizer showed significantly lower MMSE.
  • Effective Tap-Weight Shift: The tap-weight shift mechanism effectively suppresses MSE spikes when the time lag changes.
  • Low Extra Complexity: The additional computational complexity introduced by the adaptive time lag algorithm is minimal and can be considered negligible for large equalizer tap-lengths.

Limitations and Future Work

  • Analysis Scope: The authors note that a rigorous mathematical analysis of the convergence behavior of the stochastic recursion (the adaptive algorithm) is beyond the scope of the paper, though they mention it could be analyzed using probability or mean square methods.
  • Initial Guess: While the algorithm is adaptive, an initial guess for the time lag is still required. The paper suggests [(N-1)/2] as a reasonable starting point when no prior channel knowledge is available.
  • Extension to Complex Signals: The analyses are presented for real-valued signals, but the authors state that the results can be readily extended to complex signals. This would be a natural direction for future work.
  • Real-world Channel Variations: While tested on various synthetic channels, further validation on real-world, highly dynamic communication channels would be beneficial.

Summary updated

Abstract

In an adaptive equaliser, the time lag is an important parameter that significantly influences the performance. Only with the optimum time lag that corresponds to the best minimum-mean-square-error (MMSE) performance, can there be best use of the available resources. Many designs, however, choose the time lag either based on preassumption of the channel or simply based on average experience. The relation between the MMSE performance and the time lag is investigated using a new interpretation of the MMSE equaliser, and then a novel adaptive time lag algorithm is proposed based on gradient search. The proposed algorithm can converge to the optimum time lag in the mean and is verified by the numerical simulations provided.
BibTeX
@inproceedings{gong2005equalisation,
  author = {Yanmin Gong and C.F.N. Cowan},
  title = {{Equalisation with adaptive time lag}},
  booktitle = {IEE},
  year = {2005},
  doi = {10.1049/ip-com:20045305},
}
Figure