本文为《Steven M. Kay, Fundamentals of Statistical Signal Processing:Estimation Theory》一书的第2章。
1、无偏估计量
对未知参数进行估计,得到估计量。而所谓无偏估计,是指估计量的均值,等于未知参数的真值,即对未知参数 θ \theta θ,有
E ( θ ^ ) = θ , a < θ < b , (2.1) \tag{2.1} {\rm E}( \hat \theta)=\theta,\quad a<\theta<b, E(θ^)=θ,a<θ<b,(2.1)那么估计量是无偏的。
真值是确定的,估计量却是随机的,每次估计得到一个样本。因此无偏估计就是,估计量的均值等于真值。
【例2.1】AWGN中DC电平的无偏估计。
考虑观测
x [ n ] = A + w [ n ] n = 0 , 1 , … , N − 1 x[n]=A+w[n]\quad n=0,1,\ldots,N-1 x[n]=A+w[n]n=0,1,…,N−1其中 A A A是要估计的参数, w [ n ] w[n] w[n]是AWGN。参数 A A A可以取 − ∞ < A < ∞ -\infty<A<\infty −∞<A<∞上的任何值。那么, x [ n ] x[n] x[n]的一个合理估计是
A ^ = 1 N ∑ n = 0 N − 1 x [ n ] , (2.2) \tag{2.2} \hat A=\frac{1}{N}\sum_{n=0}^{N-1}x[n], A^=N1n=0∑N−1x[n],(2.2)即样本的均值。进一步,我们有
E ( A ^ ) = E [ 1 N ∑ n = 0 N − 1 x [ n ] ] = 1 N ∑ n = 0 N − 1 E ( x [ n ] ) = A \begin{aligned} {\rm E}(\hat A)&={\rm E}\left[\frac{1}{N}\sum_{n=0}^{N-1}x[n]\right]\\ &=\frac{1}{N}\sum_{n=0}^{N-1}{\rm E}(x[n])\\ &=A \end{aligned} E(A^)=E[N1n=0∑N−1x[n]]=N1n=0∑N−1E(x[n])=A因此,用样本均值作为估计量,是无偏的。
The restriction that E ( θ ^ ) = θ {\rm E}(\hat \theta)=\theta E(θ^)=θ for all θ \theta θ is an important one. Letting θ ^ = g ( x ) \hat \theta=g(\bf x) θ^=g(x), where x = [ x [ 0 ] , x [ 1 ] , … , x [ N − 1 ] ] T {\bf x}=\left[ x[0],x[1],\ldots,x[N-1]\right]^T x=[x[0],x[1],…,x[N−1]]T, it asserts that
E ( θ ^ ) = ∫ g ( x ) p ( x ; θ ) d x = θ f o r a l l θ . (2.3) \tag{2.3} {\rm E}(\hat \theta)=\int g({\bf x})p({\bf x};\theta)d{\bf x}=\theta\quad {\rm for \ all\ \theta}. E(θ^)=∫g(x)p(x;θ)dx=θfor all θ.(2.3)It is possible, however, that (2.3) may hold for some values of θ \theta θ and not others, as the next example illustrate.
Example 2.2-Biased Estimator for DC Level in White Noise
Consider again Example 2.1 but with the modified sample mean estimator
A ˇ = 1 2 N ∑ n = 0 N − 1 x [ n ] . \check{A}=\frac{1}{2N}\sum_{n=0}^{N-1}x[n]. Aˇ=2N1n=0∑N−1x[n].Then
E ( A ˇ ) = 1 2 A { = A , i f A = 0 ≠ A , i f A ≠ 0. {\rm E}(\check{A})=\frac{1}{2}A\left\{\begin{aligned} =A,\ {\rm if}A=0\\ \ne A,\ {\rm if}A\ne 0. \end{aligned}\right. E(Aˇ)=21