[Paper Review] CWCU LMMSE Estimation: Prerequisites and Properties

Mario Huemer, Oliver Lang|arXiv (Cornell University)|Dec 4, 2014

Blind Source Separation Techniques5 references3 citations

TL;DR

This paper introduces the Component-Wise Conditionally Unbiased Linear Minimum Mean Square Error (CWCU LMMSE) estimator, which relaxes the stringent global unbiasedness constraint of traditional estimators like BLUE and LMMSE by enforcing unbiasedness per parameter component. It demonstrates that under Gaussian and linear models, the CWCU LMMSE estimator leverages prior knowledge more effectively than BLUE, achieving near-LMMSE performance with conditional unbiasedness, particularly effective in channel estimation with data gaps.

ABSTRACT

The classical unbiasedness condition utilized e.g. by the best linear unbiased estimator (BLUE) is very stringent. By softening the "global" unbiasedness condition and introducing component-wise conditional unbiasedness conditions instead, the number of constraints limiting the estimator's performance can in many cases significantly be reduced. In this work we investigate the component-wise conditionally unbiased linear minimum mean square error (CWCU LMMSE) estimator for different model assumptions. The prerequisites in general differ from the ones of the LMMSE estimator. We first derive the CWCU LMMSE estimator under the jointly Gaussian assumption of the measurements and the parameters. Then we focus on the linear model and discuss the CWCU LMMSE estimator for jointly Gaussian parameters, and for mutually independent (and otherwise arbitrarily distributed) parameters, respectively. In all these cases the CWCU LMMSE estimator incorporates the prior mean and the prior covariance matrix of the parameter vector. For the remaining cases optimum linear CWCU estimators exist, but they may correspond to globally unbiased estimators that do not make use of prior statistical knowledge about the parameters. Finally, the beneficial properties of the CWCU LMMSE estimator are demonstrated with the help of a well-known channel estimation application.

Motivation & Objective

• Address the limitation of global unbiasedness in classical estimators like BLUE, which restricts the use of prior statistical knowledge.
• Develop a more flexible estimation framework by introducing component-wise conditionally unbiased (CWCU) constraints to reduce performance constraints.
• Investigate the properties and prerequisites of the CWCU LMMSE estimator under jointly Gaussian and linear model assumptions.
• Demonstrate the estimator’s superiority in practical applications such as OFDM channel estimation, especially in regions with missing measurements.
• Show that the CWCU LMMSE estimator achieves performance close to LMMSE while maintaining conditional unbiasedness, unlike BLUE.

Proposed method

• Formulate the CWCU LMMSE estimator under the jointly Gaussian assumption of parameters and measurements, deriving the estimator matrix and bias-correcting offset using conditional expectations.
• Derive the CWCU LMMSE estimator for linear models with Gaussian parameters, using the prior mean and covariance matrix to shape the estimator.
• Extend the framework to mutually independent parameters with arbitrary distributions, showing that the estimator still incorporates prior knowledge via the covariance matrix.
• Apply the estimator to a realistic OFDM channel estimation problem using a 64-subcarrier system with a 10-subcarrier gap in pilot symbols.
• Use Proposition 2 and 3 to derive the frequency-domain version of the CWCU LMMSE estimator from the time-domain estimate, preserving conditional unbiasedness.
• Compare the Bayesian MSE performance of CWCU LMMSE, LMMSE, BLUE, and trivial estimators across time and frequency domains to validate theoretical claims.

Experimental results

Research questions

• RQ1 How does relaxing global unbiasedness to component-wise conditional unbiasedness affect the performance and constraints of linear estimators?
• RQ2 What are the statistical prerequisites for the existence and optimality of the CWCU LMMSE estimator under different parameter distribution assumptions?
• RQ3 Can the CWCU LMMSE estimator achieve performance close to LMMSE while maintaining conditional unbiasedness in practical scenarios with missing data?
• RQ4 How does the CWCU LMMSE estimator compare to BLUE and LMMSE in terms of Bayesian MSE in OFDM channel estimation with pilot gaps?
• RQ5 Under what conditions does the CWCU LMMSE estimator reduce to a globally unbiased estimator, and when does it fully exploit prior knowledge?

Key findings

• The CWCU LMMSE estimator significantly reduces constraints compared to global unbiasedness, enabling better use of prior knowledge in Bayesian estimation.
• Under jointly Gaussian assumptions, the CWCU LMMSE estimator achieves optimal performance by incorporating the prior mean and covariance matrix of the parameter vector.
• In the linear model with independent parameters, the CWCU LMMSE estimator still leverages prior covariance information, even when distributions are arbitrary.
• In the OFDM channel estimation example, the CWCU LMMSE estimator achieves Bayesian MSE very close to the LMMSE estimator, especially in the data gap (subcarriers 27–37), where it outperforms BLUE by a large margin.
• The CWCU LMMSE estimator maintains conditional unbiasedness across all subcarriers, including the gap, while the BLUE suffers from a maximum Bayesian MSE of approximately 36 at subcarrier 32.
• The frequency-domain CWCU LMMSE estimator is derived via transformation using the prior covariance matrix, and it shows excellent interpolation properties across the gap, matching the LMMSE performance closely.