积化和差
cosαcosβ=12[cos(α+β)+cos(α−β)]\cos \alpha \cos \beta = \frac{1}{2}[\cos (\alpha + \beta ) + \cos (\alpha - \beta )]cosαcosβ=21[cos(α+β)+cos(α−β)]sinαcosβ=12[sin(α+β)+sin(α−β)]\sin \alpha \cos \beta = \frac{1}{2}[\sin (\alpha + \beta ) + \sin (\alpha - \beta )]sinαcosβ=21[sin(α+β)+sin(α−β)]sinαsinβ=12[cos(α−β)−cos(α+β)]\sin \alpha \sin \beta = \frac{1}{2}[\cos (\alpha - \beta ) - \cos (\alpha + \beta )]sinαsinβ=21[cos(α−β)−cos(α+β)]cosαsinβ=12[sin(α+β)−sin(α−β)]\cos \alpha \sin \beta = \frac{1}{2}[\sin (\alpha + \beta ) - \sin (\alpha - \beta )]cosαsinβ=21[sin(α+β)−sin(α−β)]
自相关与互相关
RX(t2,t1)=RX∗(t1,t2){R_X}({t_2},{t_1}) = R_X^*({t_1},{t_2})RX(t2,t1)=RX∗(t1,t2)RXY(t2,t1)=RYX∗(t1,t2){R_{XY}}({t_2},{t_1}) = R_{YX}^*({t_1},{t_2})RXY(t2,t1)=RYX∗(t1,t2)RX(t2,t1)=RX(τ),τ=t1−t2{R_X}({t_2},{t_1}) = {R_X}(\tau ),\quad \tau = {t_1} - {t_2}RX(t2,t1)=RX(τ),τ=t1−t2
矩阵微分
YYY、BBB和RRR均代表矩阵,zzz和aaa代表向量,上标T表示转置,∗*∗表示共轭,H表示共轭转置。
∂YTB∂B=Y\frac{{\partial {Y^{\mathop{\rm T}\nolimits} }B}}{{\partial B}} = Y∂B∂YTB=Y∂BTY∂B=Y\frac{{\partial {B^{\mathop{\rm T}\nolimits} }Y}}{{\partial B}} = Y∂B∂BTY=Y
规律总结: “前面”为转置,对“不转置”求导,结果为“另一个不转置”
∂BTYTYB∂B=2YTYB\frac{{\partial {B^{\mathop{\rm T}\nolimits} }{Y^{\mathop{\rm T}\nolimits} }YB}}{{\partial B}} = 2{Y^{\mathop{\rm T}\nolimits} }YB∂B∂BTYTYB=2YTYB∂BTB∂B=2B\frac{{\partial {B^{\mathop{\rm T}\nolimits} }B}}{{\partial B}} = 2B∂B∂BTB=2B∂BTWB∂B=WB+WTB\frac{{\partial {B^{\mathop{\rm T}\nolimits} }WB}}{{\partial B}} = WB + {W^{\mathop{\rm T}\nolimits} }B∂B∂BTWB=WB+WTB
特别地,WT=W{{W^{\mathop{\rm T}\nolimits} } = W}WT=W时,
∂BTWB∂B=WB+WTB=2WB\frac{{\partial {B^{\mathop{\rm T}\nolimits} }WB}}{{\partial B}} = WB + {W^{\mathop{\rm T}\nolimits} }B = 2WB∂B∂BTWB=WB+WTB=2WB
复微分
下表给出了标量函数f(w)f(\boldsymbol{w})f(w)和向量函数f(w)\boldsymbol{f}(\boldsymbol{w})f(w)关于可变向量w\boldsymbol{w}w和w∗\boldsymbol{w}^{*}w∗的复数微分结果。
函数类型 | 函数 | 变量w\boldsymbol{w}w | 变量w∗\boldsymbol{w}^{*}w∗ |
---|---|---|---|
标量f(w)f(\boldsymbol{w})f(w) | re[wHx]\operatorname{re}\left[\boldsymbol{w}^{\mathrm{H}} \boldsymbol{x}\right]re[wHx] | 12x∗\frac{1}{2}\boldsymbol{x}^{*}21x∗ | 12x\frac{1}{2}\boldsymbol{x}21x |
标量f(w)f(\boldsymbol{w})f(w) | wHx\boldsymbol{w}^{\mathrm{H}} \boldsymbol{x}wHx | 0{\bf{0}}0 | x\boldsymbol{x}x |
标量f(w)f(\boldsymbol{w})f(w) | xHw\boldsymbol{x}^{\mathrm{H}} \boldsymbol{w}xHw | x∗\boldsymbol{x}^{*}x∗ | 0{\bf{0}}0 |
标量f(w)f(\boldsymbol{w})f(w) | wHRw\boldsymbol{w}^{\mathrm{H}} \boldsymbol{R} \boldsymbol{w}wHRw | RTw∗=(RHw)∗\boldsymbol{R}^{\mathrm{T}} \boldsymbol{w}^{*}=\left(\boldsymbol{R}^{\mathrm{H}} \boldsymbol{w}\right)^{*}RTw∗=(RHw)∗ | Rw\boldsymbol{R} \boldsymbol{w}Rw |
矢量f(w)\boldsymbol{f}(\boldsymbol{w})f(w) | H1w+H2w∗\boldsymbol{H}_{1} \boldsymbol{w}+\boldsymbol{H}_{2} \boldsymbol{w}^{*}H1w+H2w∗ | H1T\boldsymbol{H}_{1}^{\mathrm{T}}H1T | H2T\boldsymbol{H}_{2}^{\mathrm{T}}H2T |
一个计算小技巧
已知基向量两两正交
∫−∞∞fm(t)fn∗(t)dt=δ(m−n)\int_{ - \infty }^\infty {{f_m}(t)f_n^*(t)dt} = \delta (m - n)∫−∞∞fm(t)fn∗(t)dt=δ(m−n)
s(t){s(t)}s(t)可由基向量线性组合近似
s^(t)=∑k=1Kskfk(t)\hat s(t) = \sum\limits_{k = 1}^K {{s_k}{f_k}(t)}s^(t)=k=1∑Kskfk(t)
由误差与基向量正交,有
⟨s(t)−s^(t),fn(t)⟩=0⇒sn=⟨s(t),fn(t)⟩\left\langle {s(t) - \hat s(t),{f_n}(t)} \right\rangle = 0 \Rightarrow {s_n} = \left\langle {s(t),{f_n}(t)} \right\rangle⟨s(t)−s^(t),fn(t)⟩=0⇒sn=⟨s(t),fn(t)⟩
则误差的二范数为
εe=∫−∞∞(s(t)−s^(t))(s(t)−s^(t))∗dt{\varepsilon _e} = \int_{ - \infty }^\infty {(s(t) - \hat s(t)){{(s(t) - \hat s(t))}^*}dt}εe=∫−∞∞(s(t)−s^(t))(s(t)−s^(t))∗dt
=∫−∞∞∣s(t)∣2dt−∫−∞∞∑k=1Kskfk(t)⋅s∗(t)dt−⟨s(t)−s^(t),s^∗(t)⟩= \int_{ - \infty }^\infty {|s(t){|^2}dt} - \int_{ - \infty }^\infty {\sum\limits_{k = 1}^K {{s_k}{f_k}(t)} \cdot {s^*}(t)dt}- \left\langle {s(t) - \hat s(t),{{\hat s}^*}(t)} \right\rangle=∫−∞∞∣s(t)∣2dt−∫−∞∞k=1∑Kskfk(t)⋅s∗(t)dt−⟨s(t)−s^(t),s^∗(t)⟩
=∫−∞∞∣s(t)∣2dt−∑k=1Ksk[∫−∞∞s(t)fk∗(t)dt]∗= \int_{ - \infty }^\infty {|s(t){|^2}dt} - \sum\limits_{k = 1}^K {{s_k}{{[\int_{ - \infty }^\infty {s(t)f_k^*(t)dt} ]}^*}}=∫−∞∞∣s(t)∣2dt−k=1∑Ksk[∫−∞∞s(t)fk∗(t)dt]∗
=∫−∞∞∣s(t)∣2dt−∑k=1Ksksk∗= \int_{ - \infty }^\infty {|s(t){|^2}dt} - \sum\limits_{k = 1}^K {{s_k}s_k^*}=∫−∞∞∣s(t)∣2dt−k=1∑Ksksk∗
=∫−∞∞∣s(t)∣2dt−∑k=1K∣sk∣2= \int_{ - \infty }^\infty {|s(t){|^2}dt} - \sum\limits_{k = 1}^K {|{s_k}{|^2}}=∫−∞∞∣s(t)∣2dt−k=1∑K∣sk∣2