The zero-forcing precoder aims to make the estimate of the transmitted signal x unbiased after detection by satisfying E{f*x̃|x} = x. This leads to the constraint that fGHP = I. The zero-forcing optimization problem is then to minimize the mean-squared error E‖f*x̃ - x‖^2 subject to this constraint and a transmit energy constraint. Solving the optimization problem using Lagrange multipliers yields the zero-forcing precoding matrix in (14), which depends inversely on the channel covariance matrix and directly on the transmit energy.
The zero-forcing precoder aims to make the estimate of the transmitted signal x unbiased after detection by satisfying E{f*x̃|x} = x. This leads to the constraint that fGHP = I. The zero-forcing optimization problem is then to minimize the mean-squared error E‖f*x̃ - x‖^2 subject to this constraint and a transmit energy constraint. Solving the optimization problem using Lagrange multipliers yields the zero-forcing precoding matrix in (14), which depends inversely on the channel covariance matrix and directly on the transmit energy.
The zero-forcing precoder aims to make the estimate of the transmitted signal x unbiased after detection by satisfying E{f*x̃|x} = x. This leads to the constraint that fGHP = I. The zero-forcing optimization problem is then to minimize the mean-squared error E‖f*x̃ - x‖^2 subject to this constraint and a transmit energy constraint. Solving the optimization problem using Lagrange multipliers yields the zero-forcing precoding matrix in (14), which depends inversely on the channel covariance matrix and directly on the transmit energy.
The zero-forcing precoder aims to make the estimate of the transmitted signal x unbiased after detection by satisfying E{f*x̃|x} = x. This leads to the constraint that fGHP = I. The zero-forcing optimization problem is then to minimize the mean-squared error E‖f*x̃ - x‖^2 subject to this constraint and a transmit energy constraint. Solving the optimization problem using Lagrange multipliers yields the zero-forcing precoding matrix in (14), which depends inversely on the channel covariance matrix and directly on the transmit energy.
It is considered that the MIMO system consists of a linear precoder with
precoding matrix P at the transmitter and a linear equalizer represented by the matrix G. The output signal of the detector is described by
x̃ = G (HPx + w) , (1)
where H is the MIMO channel, x is the input signal and
n w iso the additive noise. The input signal x has the covariance matrix E xxH = Cx . The n o noise w has the covariance matrix E wwH = Cw . Beside the receive filter G we consider a scaling factor f. The zero-forcing precoder is characterized by a design criterion which is unbiased in terms of E {f x̃|x} = x. In the context of the given system model the unbiasedness can be expressed as:
E {f x̃|x} = x (2) fGHPx = x (3) fGHP = I. (4)
The unbiasedness can be used as an additional constraint in the MMSE pre-
coding optimization problem, such that the zero-forcing optimization prob- lem can be described as: n o PZF = argmin E kf x̃ − xk22 (5) P,f
s.t. fGHP = I n o tr PCx PH = ETx ,
where we assume that the transmit energy is equal to ETx . With the con- straint (4) the objective function can be reduced, such that the optimization problem is as follows n o PZF = argmin E k−fGwk22 (6) P,f
s.t. fGHP = I n o tr PCx PH = ETx , With this, the Lagrangian function is given by n o 1 L (P, f, Λ, λ) =f 2 tr GCw GH + tr {Λ (fGHP − I)} (7) 2 1 n H H H o + tr fP H G − I ΛH 2 n o + λ tr PCx PH − ETx n o 1 =f 2 tr GCw GH + tr {Λ (fGHP − I)} 2 1 n H H H Ho 1 n Ho + tr fH G Λ P − tr Λ 2 n o 2 H + λ tr PCx P − ETx . The derivative with respect to P∗ is given by ∂L (P, f, Λ, λ) f = HH GH ΛH + λPCx . (8) ∂P∗ 2 Equating it to zero yields f H H H −1 P=− H G Λ Cx . (9) 2λ Using (9) together with the zero-forcing constraint (4) yields −f H H H −1 fGH H G Λ Cx = I (10) 2λ −2λ −1 ΛH = 2 GHHH GH Cx . f Replacing ΛH in (9) yields 1 −1 P = HH GH GHHH GH . (11) f Using (11) in the transmit energy constraint yields 1 −1 −1 ETx = tr 2 HH GH GHHH GH Cx GHHH GH GH , (12) f which can be rearranged such that v n o u tr Cx (GHHH GH )−1 u t f= . (13) ETx Finally the zero-forcing precoding matrix is given by: v u ETx −1 PZF = t o HH GH GHHH GH . (14) u n −1 tr Cx (GHHH GH ) References [Joham] M. Joham, W. Utschick and J. A. Nossek, “Linear transmit processing in MIMO communications systems,” in IEEE Transactions on Signal Processing, vol. 53, no. 8, pp. 2700-2712, Aug. 2005.
