115 年 國立中正大學通訊工程學系碩士班通訊甲組《通訊理論》

📄 試題原卷 免費註冊後即可對照原始考卷 PDF免費註冊

第 1 題25 分

  1. (25%) Consider the transmission of an analog message signal m(t) over a noiseless channel via an impulse train sampler, i.e., the transmitted signal is given by
    x(t)=∑n=−∞∞m(nT)δ(t–nT)x(t) = \sum_{n=-\infty}^{\infty} m(nT)\delta(t – nT)
    where T>0T > 0 denotes the sampling period and δ(t)\delta(t) is the Dirac delta function. Upon receiving x(t)x(t), a reconstruction filter with frequency response
H(f)={T,∣f∣≤12T0,otherwiseH(f) = \begin{cases} T, & |f| \le \frac{1}{2T} \\ 0, & \text{otherwise} \end{cases}

is used to recover the message signal; denoted by m~(t)\tilde{m}(t) the filter output.

(a) (10%) Derive an expression for m~(t)\tilde{m}(t) in terms of the samples {m(nT):n∈Z}\{m(nT) : n \in \mathbb{Z}\}.
(b) (5%) Assume that m(t)m(t) is bandlimited to B(Hz)B(\text{Hz}) with B<1/(2T)B < 1/(2T). Determine whether m(t)m(t) can be perfectly reconstructed. Explain your answer.
(c) (10%) If only the signal values at the sampling instants t=nTt = nT are of interest, that is, we require
m~(nT)=m(nT),∀n∈Z,\tilde{m}(nT) = m(nT), \quad \forall n \in \mathbb{Z},
is the bandlimitedness assumption on m(t)m(t) necessary under this requirement?

登入後即可作答並保存紀錄。

這一題的完整詳解

核心觀念

本題考查理想脈衝取樣、取樣訊號的頻譜週期延拓,以及理想低通濾波器所形成的 sinc 插值公式。

採用傅立葉轉換定義:

M(f)=∫−∞∞m(t)e−j2πft dtM(f)=\int_{-\infty}^{\infty}m(t)e^{-j2\pi ft}\,dt

並使用正規化 sinc 函數:

sinc⁡(u)=sin⁡(πu)πu.\operatorname{sinc}(u)=\frac{\sin(\pi u)}{\pi u}.

取樣後的訊號為

x(t)=∑n=−∞∞m(nT)δ(t−nT).x(t)=\sum_{n=-\infty}^{\infty}m(nT)\delta(t-nT).

理想重建濾波器的頻率響應為通帶增益 TT、截止頻率 1/(2T)1/(2T) 的低通濾波器。


解題方法

重建濾波器輸出等於取樣訊號與濾波器脈衝響應的卷積:

m~(t)=x(t)∗h(t).\tilde m(t)=x(t)*h(t).

先由 H(f)H(f) 求出 h(t)h(t),再利用脈衝函數的平移與取樣性質,即可得到時域插值公式。


(a) 重建訊號的表示式

由反傅立葉轉換,

h(t)=∫−∞∞H(f)ej2πft df.h(t)=\int_{-\infty}^{\infty}H(f)e^{j2\pi ft}\,df.

代入題目給定的 H(f)H(f):

h(t)=T∫−1/(2T)1/(2T)ej2πft df.h(t) = T\int_{-1/(2T)}^{1/(2T)}e^{j2\pi ft}\,df.

計算得

h(t)=T[ej2πftj2πt]−1/(2T)1/(2T)=Tsin⁡(πt/T)πt=sin⁡(πt/T)πt/T=sinc⁡(tT).\begin{aligned} h(t) &= T\left[ \frac{e^{j2\pi ft}}{j2\pi t} \right]_{-1/(2T)}^{1/(2T)}\\ &= T\frac{\sin(\pi t/T)}{\pi t}\\ &= \frac{\sin(\pi t/T)}{\pi t/T}\\ &= \operatorname{sinc}\left(\frac{t}{T}\right). \end{aligned}

因此,

m~(t)=x(t)∗h(t)=∑n=−∞∞m(nT)δ(t−nT)∗sinc⁡(tT).\tilde m(t) = x(t)*h(t) = \sum_{n=-\infty}^{\infty} m(nT)\delta(t-nT) * \operatorname{sinc}\left(\frac{t}{T}\right).

利用

δ(t−nT)∗h(t)=h(t−nT),\delta(t-nT)*h(t)=h(t-nT),

可得

m~(t)=∑n=−∞∞m(nT)sinc⁡(t−nTT)\boxed{ \tilde m(t) = \sum_{n=-\infty}^{\infty} m(nT) \operatorname{sinc}\left(\frac{t-nT}{T}\right) }

也可寫成

m~(t)=∑n=−∞∞m(nT)sin⁡[π(tT−n)]π(tT−n).\boxed{ \tilde m(t) = \sum_{n=-\infty}^{\infty} m(nT) \frac{\sin\left[\pi\left(\frac{t}{T}-n\right)\right]} {\pi\left(\frac{t}{T}-n\right)} }.

這就是 Shannon sampling interpolation formula。每一個樣本值 m(nT)m(nT) 都乘上一個以 t=nTt=nT 為中心的 sinc 函數,再將所有 sinc 函數相加。


(b) 是否能完美重建 m(t)m(t)?

頻域分析

脈衝取樣器的頻域結果為

X(f)=1T∑k=−∞∞M(f−kT).X(f)=\frac{1}{T}\sum_{k=-\infty}^{\infty}M\left(f-\frac{k}{T}\right).

這表示原始訊號頻譜 M(f)M(f) 會以 1/T1/T Hz 為週期重複出現,且每個頻譜副本的振幅乘上 1/T1/T。

輸出頻譜為

M~(f)=H(f)X(f).\tilde M(f)=H(f)X(f).

由於

H(f)=T,∣f∣≤12T,H(f)=T,\qquad |f|\le \frac{1}{2T},

在重建濾波器通帶內,

M~(f)=T⋅1T∑k=−∞∞M(f−kT).\tilde M(f) = T\cdot \frac{1}{T} \sum_{k=-\infty}^{\infty} M\left(f-\frac{k}{T}\right).

題目給定

M(f)=0,∣f∣>B,M(f)=0,\qquad |f|>B,

且

B<12T.B<\frac{1}{2T}.
🔒

後續完整解題步驟與【答案】

免費註冊,享三天全站完整詳解閱覽。

免費註冊

第 2 題25 分

  1. (25%) Consider a conventional waveform channel y(t)=x(t)+n(t)y(t) = x(t)+n(t), where the noise process n(t)n(t) is described by a sequence of random pulse train given by
    n(t)=∑i=−∞∞nip(t−iT+δ)n(t) = \sum_{i=-\infty}^{\infty} n_i p(t-iT + \delta)
    Here, the waveform p(t)p(t) is a rectangular pulse with height 1/T1/\sqrt{T} on [0,T][0, T] and 00 elsewhere, the parameter δ∈[0,T)\delta \in [0, T) is a random time offset uniformly distributed on [0,T)[0, T), and the weighting coefficients nin_i's are independent and identically distributed Gaussian random variables with zero-mean and variance σn2\sigma_n^2. Assume that δ\delta is independent of nin_i's.

(a) (8%) Is the noise process n(t)n(t) wide-sense stationary (WSS)? Explain your answer.
(b) (6%) Describe the limiting behavior of the autocorrelation function and power spectral density of n(t)n(t) as T→0T \rightarrow 0.
(c) (4%) Let w(t)w(t) be the output of the ideal low-pass filter of bandwidth B(Hz)B(\text{Hz}) and height 11 with input n(t)n(t). Find the output variance of w(t)w(t) in the limit as T→0T \rightarrow 0.
(d) (7%) As T→0T \rightarrow 0, the noise process n(t)n(t) becomes Gaussian since any finite number of samples of n(t)n(t) are jointly Gaussian. Discuss whether w(t)w(t) defined in part (c) is also a Gaussian process and whether it is WSS. Explain your answer.

登入後即可作答並保存紀錄。

這一題的完整詳解

核心觀念

本題探討數位通訊與隨機程序中的經典模型——隨機脈衝序列(Random Pulse Train / PAM Signal) 的統計特性分析。核心觀念與定理包含:

  1. 隨機時延與廣義平穩性(WSS):一般的脈衝序列具有週期平穩性(Cyclostationary)。引入均勻分佈於 [0,T)[0, T) 的隨機時延 δ\delta(Random Time Offset / Jitter)後,可將週期非平穩過程轉化為嚴格的廣義平穩過程(Wide-Sense Stationary, WSS)。
  2. 全期望公式與區間拼接(Conditioning and Variable Substitution):計算隨機程序的平均值 E[n(t)]E[n(t)] 與自相關函數 Rn(t,t+τ)=E[n(t)n(t+τ)]R_n(t, t+\tau) = E[n(t)n(t+\tau)],透過條件期望值 E[⋅∣δ]E[\cdot \mid \delta] 與變數變換 θ=t−iT+δ\theta = t - iT + \delta,將離散級數和拼接為全實數軸連續積分。
  3. 白雜訊(White Noise)極限行為:當脈衝寬度 T→0T \to 0 且能量正規化時,自相關函數 Rn(τ)R_n(\tau) 趨近於狄拉克函數(Dirac Delta Function)δ(τ)\delta(\tau),其功率譜密度(Power Spectral Density, PSD)Sn(f)S_n(f) 趨近於全頻帶均勻分佈之常數。
  4. 線性時不變(LTI)系統之輸出特性:
    • 功率譜密度關係:Sw(f)=Sn(f)∣H(f)∣2S_w(f) = S_n(f)|H(f)|^2
    • 輸出方差(Variance):Var⁡(w(t))=Rw(0)=∫−∞∞Sw(f) df\operatorname{Var}(w(t)) = R_w(0) = \int_{-\infty}^{\infty} S_w(f) \, df
  5. 高斯程序(Gaussian Process)之線性不變性:高斯隨機程序通過任意穩定 LTI 系統後,其輸出仍為高斯隨機程序;若輸入為 WSS,則輸出亦必為 WSS。

詳細解題步驟與推導

(a) 判斷 n(t)n(t) 是否為廣義平穩過程(WSS)

Step 1:檢驗平均值 E[n(t)]E[n(t)]
由於 nin_i 與 δ\delta 相互獨立,且 E[ni]=0E[n_i] = 0,根據全期望公式:
E[n(t)]=Eδ[∑i=−∞∞E[ni]⋅p(t−iT+δ)]=Eδ[∑i=−∞∞0⋅p(t−iT+δ)]=0E[n(t)] = E_{\delta}\left[ \sum_{i=-\infty}^{\infty} E[n_i] \cdot p(t - iT + \delta) \right] = E_{\delta}\left[ \sum_{i=-\infty}^{\infty} 0 \cdot p(t - iT + \delta) \right] = 0
平均值為與時間 tt 無關的常數 00。

Step 2:計算自相關函數 Rn(t,t+τ)=E[n(t)n(t+τ)]R_n(t, t+\tau) = E[n(t)n(t+\tau)]
給定隨機變數 δ\delta 的條件下,由於 nin_i 彼此獨立且同分佈(i.i.d.),其二階互相關滿足 E[ninj]=σn2δijE[n_i n_j] = \sigma_n^2 \delta_{ij}(其中 δij\delta_{ij} 為 Kronecker delta):
E[n(t)n(t+τ)∣δ]=∑i=−∞∞∑j=−∞∞E[ninj]⋅p(t−iT+δ)p(t+τ−jT+δ)E[n(t)n(t+\tau) \mid \delta] = \sum_{i=-\infty}^{\infty} \sum_{j=-\infty}^{\infty} E[n_i n_j] \cdot p(t - iT + \delta) p(t + \tau - jT + \delta)
=σn2∑i=−∞∞p(t−iT+δ)p(t+τ−iT+δ)= \sigma_n^2 \sum_{i=-\infty}^{\infty} p(t - iT + \delta) p(t + \tau - iT + \delta)

對 δ∼Uniform[0,T)\delta \sim \text{Uniform}[0, T) 取期望值(其機率密度函數為 fδ(δ)=1T, δ∈[0,T)f_\delta(\delta) = \frac{1}{T}, \, \delta \in [0, T)):
Rn(t,t+τ)=σn2T∫0T∑i=−∞∞p(t−iT+δ)p(t+τ−iT+δ) dδR_n(t, t+\tau) = \frac{\sigma_n^2}{T} \int_{0}^{T} \sum_{i=-\infty}^{\infty} p(t - iT + \delta) p(t + \tau - iT + \delta) \, d\delta

令變數變換 θ=t−iT+δ\theta = t - iT + \delta,則 dθ=dδd\theta = d\delta。當 δ\delta 從 00 積分到 TT 時,θ\theta 的積分區間為 [t−iT, t−(i−1)T)[t - iT, \, t - (i-1)T)。將所有 i∈Zi \in \mathbb{Z} 的區間無縫拼接後,覆蓋整個實數軸 (−∞,∞)(-\infty, \infty):
Rn(t,t+τ)=σn2T∑i=−∞∞∫t−iTt−(i−1)Tp(θ)p(θ+τ) dθ=σn2T∫−∞∞p(θ)p(θ+τ) dθR_n(t, t+\tau) = \frac{\sigma_n^2}{T} \sum_{i=-\infty}^{\infty} \int_{t-iT}^{t-(i-1)T} p(\theta) p(\theta + \tau) \, d\theta = \frac{\sigma_n^2}{T} \int_{-\infty}^{\infty} p(\theta) p(\theta + \tau) \, d\theta

定義脈衝 p(t)p(t) 自身的自相關函數為 Rp(τ)=∫−∞∞p(θ)p(θ+τ) dθR_p(\tau) = \int_{-\infty}^{\infty} p(\theta) p(\theta + \tau) \, d\theta。
已知 p(t)=1Trect(t−T/2T)p(t) = \frac{1}{\sqrt{T}} \text{rect}\left(\frac{t - T/2}{T}\right),為寬度 TT、高度 1/T1/\sqrt{T} 之矩形脈衝,其自身卷積與積分結果為對稱三角脈衝:
Rp(τ)={1−∣τ∣T,∣τ∣≤T0,∣τ∣>T=tri⁡(τT)R_p(\tau) = \begin{cases} 1 - \frac{|\tau|}{T}, & |\tau| \le T \\ 0, & |\tau| > T \end{cases} = \operatorname{tri}\left(\frac{\tau}{T}\right)

代入得自相關函數為:
Rn(τ)=σn2Ttri⁡(τT)={σn2T(1−∣τ∣T),∣τ∣≤T0,∣τ∣>TR_n(\tau) = \frac{\sigma_n^2}{T} \operatorname{tri}\left(\frac{\tau}{T}\right) = \begin{cases} \frac{\sigma_n^2}{T} \left(1 - \frac{|\tau|}{T}\right), & |\tau| \le T \\ 0, & |\tau| > T \end{cases}

結論:
因為 E[n(t)]=0E[n(t)] = 0(常數)且 Rn(t,t+τ)=Rn(τ)R_n(t, t+\tau) = R_n(\tau) 僅取決於時間差 τ\tau,與絕對時間 tt 無關,故雜訊程序 n(t)n(t) 是廣義平穩程序(WSS)。


(b) 當 T→0T \rightarrow 0 時,Rn(τ)R_n(\tau) 與 Sn(f)S_n(f) 的極限行為

1. 自相關函數 Rn(τ)R_n(\tau) 的極限行為:
自相關函數為 Rn(τ)=σn2⋅[1Ttri⁡(τT)]R_n(\tau) = \sigma_n^2 \cdot \left[ \frac{1}{T} \operatorname{tri}\left(\frac{\tau}{T}\right) \right]。
函數 gT(τ)=1Ttri⁡(τT)g_T(\tau) = \frac{1}{T} \operatorname{tri}\left(\frac{\tau}{T}\right) 具備以下性質:

  • 寬度為 2T→T→002T \xrightarrow{T \to 0} 0
  • 峰值高度為 1T→T→0∞\frac{1}{T} \xrightarrow{T \to 0} \infty
  • 總面積為 ∫−∞∞gT(τ) dτ=1T⋅(12×2T×1)=1\int_{-\infty}^{\infty} g_T(\tau) \, d\tau = \frac{1}{T} \cdot \left(\frac{1}{2} \times 2T \times 1\right) = 1

依據廣義函數(Delta Sequences)定義,當 T→0T \to 0 時,gT(τ)→δ(τ)g_T(\tau) \to \delta(\tau)。
因此:

🔒

後續完整解題步驟與【答案】

免費註冊,享三天全站完整詳解閱覽。

免費註冊

第 3 題20 分

  1. (20%) Consider that a single-tone message signal m(t)=Amcos⁡(2πfmt)m(t) = A_m \cos(2\pi f_m t) is used to generate the following modulated signal with carrier amplitude AcA_c and frequency fc(≫fm)f_c (\gg f_m):
    s(t)=12AmAccos⁡[2π(fc+fm)t]+12(1−Am)Accos⁡[2π(fc−fm)t]s(t) = \frac{1}{2}A_m A_c \cos[2\pi(f_c+f_m)t] + \frac{1}{2}(1-A_m)A_c \cos[2\pi(f_c-f_m)t]
    where Am∈[0,1]A_m \in [0, 1] denotes the attenuation factor applied to the upper side frequency.

(a) (4%) Find the in-phase and quadrature components of s(t)s(t).
(b) (4%) Identify the modulation type represented by s(t)s(t) for Am=0A_m = 0, Am=0.5A_m = 0.5, Am=1A_m = 1, and for other values of aa. (Note: the problem statement uses AmA_m for attenuation factor, but asks for values of aa. Assuming aa refers to AmA_m).
(c) (8%) Assume that the signal s(t)s(t) is, plus the carrier Accos⁡(2πfct)A_c \cos(2\pi f_c t), is sent and passed through an envelope detector. In the absence of channel noise, determine the envelope distortion introduced by the quadrature component in the detector output.
(d) (4%) Based on your result in part (c), determine the values of AmA_m for which the distortion is eliminated and the values for which it is maximized. Justify your answer.

登入後即可作答並保存紀錄。

這一題的完整詳解

核心觀念

本題考查:

  1. 利用和差公式將上下邊頻分解成同相(in-phase)與正交(quadrature)分量。
  2. 由上下邊頻的相對振幅判斷調變種類。
  3. 包絡檢波器實際量測的是射頻訊號的包絡:
    e(t)=xI2(t)+xQ2(t)e(t)=\sqrt{x_I^2(t)+x_Q^2(t)}
  4. 正交分量無法直接以線性方式通過包絡檢波器,會以平方項形式造成包絡失真。

令

ωc=2πfc,ωm=2πfm\omega_c=2\pi f_c,\qquad \omega_m=2\pi f_m

並將 attenuation factor 統一記為 AmA_m。


(a)同相與正交分量

原訊號為

s(t)=12AmAccos⁡[(ωc+ωm)t]+12(1−Am)Accos⁡[(ωc−ωm)t].s(t)=\frac{1}{2}A_mA_c\cos[(\omega_c+\omega_m)t] +\frac{1}{2}(1-A_m)A_c\cos[(\omega_c-\omega_m)t].

利用

cos⁡(ωc+ωm)t=cos⁡ωctcos⁡ωmt−sin⁡ωctsin⁡ωmt\cos(\omega_c+\omega_m)t =\cos\omega_ct\cos\omega_mt-\sin\omega_ct\sin\omega_mt

以及

cos⁡(ωc−ωm)t=cos⁡ωctcos⁡ωmt+sin⁡ωctsin⁡ωmt,\cos(\omega_c-\omega_m)t =\cos\omega_ct\cos\omega_mt+\sin\omega_ct\sin\omega_mt,

可得

s(t)=Ac2Am(cos⁡ωctcos⁡ωmt−sin⁡ωctsin⁡ωmt)+Ac2(1−Am)(cos⁡ωctcos⁡ωmt+sin⁡ωctsin⁡ωmt).\begin{aligned} s(t) ={}&\frac{A_c}{2}A_m \left(\cos\omega_ct\cos\omega_mt-\sin\omega_ct\sin\omega_mt\right)\\ &+\frac{A_c}{2}(1-A_m) \left(\cos\omega_ct\cos\omega_mt+\sin\omega_ct\sin\omega_mt\right). \end{aligned}

整理 cos⁡ωct\cos\omega_ct 與 sin⁡ωct\sin\omega_ct 的係數:

s(t)=Ac2cos⁡ωmtcos⁡ωct+Ac2(1−2Am)sin⁡ωmtsin⁡ωct.s(t)=\frac{A_c}{2}\cos\omega_mt\cos\omega_ct +\frac{A_c}{2}(1-2A_m)\sin\omega_mt\sin\omega_ct.

採用通訊系統常見表示式

s(t)=Ac[I(t)cos⁡ωct−Q(t)sin⁡ωct],s(t)=A_c\left[I(t)\cos\omega_ct-Q(t)\sin\omega_ct\right],

因此

I(t)=12cos⁡ωmtI(t)=\frac{1}{2}\cos\omega_mt

且

Q(t)=2Am−12sin⁡ωmt.Q(t)=\frac{2A_m-1}{2}\sin\omega_mt.

所以:

  • 同相分量:
I(t)=12cos⁡(2πfmt)\boxed{I(t)=\frac{1}{2}\cos(2\pi f_mt)}
  • 正交分量:
Q(t)=2Am−12sin⁡(2πfmt)\boxed{Q(t)=\frac{2A_m-1}{2}\sin(2\pi f_mt)}

若採用 s(t)=Ac[I(t)cos⁡ωct+Q(t)sin⁡ωct]s(t)=A_c[I(t)\cos\omega_ct+Q(t)\sin\omega_ct] 的符號慣例,正交分量會寫成 1−2Am2sin⁡ωmt\frac{1-2A_m}{2}\sin\omega_mt;物理結果完全相同。


(b)調變種類判斷

上下邊頻的振幅分別為

AUSB=12AmAcA_{\mathrm{USB}}=\frac{1}{2}A_mA_c

以及

ALSB=12(1−Am)Ac.A_{\mathrm{LSB}}=\frac{1}{2}(1-A_m)A_c.

題目中的訊號沒有獨立載波,因此屬於 suppressed-carrier 類型。

Am=0A_m=0

此時

AUSB=0,ALSB=12Ac.A_{\mathrm{USB}}=0,\qquad A_{\mathrm{LSB}}=\frac{1}{2}A_c.

只剩下低側頻,因此為:

下邊帶抑制載波調變(LSB-SC)\boxed{\text{下邊帶抑制載波調變(LSB-SC)}}

Am=0.5A_m=0.5

此時

AUSB=ALSB=14Ac.A_{\mathrm{USB}}=A_{\mathrm{LSB}}=\frac{1}{4}A_c.

上下邊頻振幅相等,且沒有載波,因此為:

雙邊帶抑制載波調變(DSB-SC)\boxed{\text{雙邊帶抑制載波調變(DSB-SC)}}

此時由(a)可見 Q(t)=0Q(t)=0,訊號完全是同相分量。

Am=1A_m=1

此時

AUSB=12Ac,ALSB=0.A_{\mathrm{USB}}=\frac{1}{2}A_c,\qquad A_{\mathrm{LSB}}=0.

只剩下高側頻,因此為:

上邊帶抑制載波調變(USB-SC)\boxed{\text{上邊帶抑制載波調變(USB-SC)}}

其他情況

當

0<Am<1,Am≠0.5,0<A_m<1,\qquad A_m\neq 0.5,

上下邊頻同時存在,但振幅不相等,故為:

非對稱雙邊帶抑制載波調變\boxed{\text{非對稱雙邊帶抑制載波調變}}

也可稱為 asymmetric DSB-SC。


(c)包絡檢波器造成的失真

題目指定將載波

Accos⁡ωctA_c\cos\omega_ct

與 s(t)s(t) 相加,因此總訊號為

x(t)=Accos⁡ωct+s(t).x(t)=A_c\cos\omega_ct+s(t).

代入(a)的結果:

🔒

後續完整解題步驟與【答案】

免費註冊,享三天全站完整詳解閱覽。

免費註冊

第 4 題30 分

  1. (30%) Consider the discrete-time channel
    y=h∗x+ny = h * x + n
    where "*" is the convolution operator, h=[1,−1,1]h = [1, -1, 1] denotes the channel impulse response, x=[x1,x2]x = [x_1, x_2] represents the transmitted vector with two independent and equiprobable signals selected from the set {+1,−1}\{+1, -1\}, and the noise vector nn consists of independent and identically distributed Gaussian random variables with zero-mean and variance σn2\sigma_n^2. Assume full linear convolution with zero padding. Answer the following questions:

(a) (2%) Express yy in matrix form y=Hx+ny = Hx+n by specifying the channel matrix HH.
(b) (5%) Based on the representation in part (a), derive the maximum-likelihood (ML) decision rule for estimating x1x_1 from yy, given HH and the noise statistics.
(c) (5%) Assume that σn2\sigma_n^2 is sufficiently small, derive an approximation of the ML decoding error probability for x1x_1 under the decision rule obtained in part (b). Express your result in terms of the Q-function, where Q(t)=∫t∞12πe−u2/2duQ(t) = \int_t^\infty \frac{1}{\sqrt{2\pi}} e^{-u^2/2} du.
(d) (5%) In the absence of noise, design a linear left-inverse system WW that minimizes the mean-squared error Ex[∥Wy−x∣∣2]E_x [\| Wy - x ||^2], where the expectation is taken over all equiprobable transmitted vectors xx and ∣∣⋅∣∣|| \cdot || denotes the Euclidean norm.
(e) (5%) Let y~=Wy\tilde{y} = Wy denote the output vector of the inverse system in part (d). Determine the probability distribution of the effective noise vector n~=Wn\tilde{n} = Wn.
(f) (5%) Suppose that the optimal decoding of x1x_1 is derived based on the scalar observation y1=x1+n~1y_1 = x_1 + \tilde{n}_1. What is the decoding error probability for sufficiently small σn2\sigma_n^2? Express your result in terms of the Q-function defined in part (c).
(g) (3%) Even though both decoders are optimal under their respective models, one based on yy and the other based on y~\tilde{y}, their performance is not the same. Explain why.

登入後即可作答並保存紀錄。

這一題的完整詳解

核心觀念

本題綜合考查:

  • 線性卷積的矩陣表示。
  • AWGN 通道下的最大概似決策。
  • 高訊噪比下,以訊號集合間最小歐氏距離近似錯誤率。
  • Moore–Penrose 線性左逆矩陣。
  • 高斯雜訊經線性轉換後的分布。
  • 雜訊增強與充分統計量的概念。

令

h=[1,−1,1],x=[x1,x2]T.h=[1,-1,1],\qquad x=[x_1,x_2]^T.

採用完整線性卷積並補零,因此輸出長度為 3+2−1=43+2-1=4。


(a) 矩陣形式

完整卷積為

y1=x1+n1,y2=−x1+x2+n2,y3=x1−x2+n3,y4=x2+n4.\begin{aligned} y_1&=x_1+n_1,\\ y_2&=-x_1+x_2+n_2,\\ y_3&=x_1-x_2+n_3,\\ y_4&=x_2+n_4. \end{aligned}

因此

y=Hx+n,y=Hx+n,

其中

H=[10−111−101].H= \begin{bmatrix} 1&0\\ -1&1\\ 1&-1\\ 0&1 \end{bmatrix}.

令 HH 的兩個 column vector 為

a=[1−110],b=[01−11],a= \begin{bmatrix} 1\\-1\\1\\0 \end{bmatrix}, \qquad b= \begin{bmatrix} 0\\1\\-1\\1 \end{bmatrix},

則

y=ax1+bx2+n.y=ax_1+bx_2+n.

(b) x1x_1 的最大概似決策規則

解題方法

因為 x2x_2 是未知干擾訊號,估計 x1x_1 時必須將 x2x_2 邊際化。由於 x1,x2x_1,x_2 獨立且等機率,最大概似規則為

x^1=arg⁡max⁡s∈{+1,−1}p(y∣x1=s),\hat{x}_1 = \arg\max_{s\in\{+1,-1\}}p(y\mid x_1=s),

其中

p(y∣x1=s)=12∑t∈{+1,−1}p(y∣x1=s,x2=t).p(y\mid x_1=s) = \frac{1}{2} \sum_{t\in\{+1,-1\}} p(y\mid x_1=s,x_2=t).

AWGN 的條件機率密度為

p(y∣x1=s,x2=t)∝exp⁡(−∥y−sa−tb∥22σn2).p(y\mid x_1=s,x_2=t) \propto \exp\left( -\frac{\|y-sa-tb\|^2}{2\sigma_n^2} \right).

因此可寫成

x^1=arg⁡max⁡s∈{+1,−1}∑t=±1exp⁡(−∥y−sa−tb∥22σn2).\hat{x}_1 = \arg\max_{s\in\{+1,-1\}} \sum_{t=\pm1} \exp\left( -\frac{\|y-sa-tb\|^2}{2\sigma_n^2} \right).

計算所需內積:

aTa=3,bTb=3,aTb=−2.a^Ta=3,\qquad b^Tb=3,\qquad a^Tb=-2.

令

A=aTy,B=bTy.A=a^Ty,\qquad B=b^Ty.

則

∥y−sa−tb∥2=∥y∥2+6+2st−2sA−2tB.\|y-sa-tb\|^2 = \|y\|^2+6+2st-2sA-2tB.

去除與 ss 無關的共同項後,

p(y∣x1=s)∝exp⁡(sAσn2)∑t=±1exp⁡(t(B+2s)σn2).p(y\mid x_1=s) \propto \exp\left(\frac{sA}{\sigma_n^2}\right) \sum_{t=\pm1} \exp\left(\frac{t(B+2s)}{\sigma_n^2}\right).

利用

ez+e−z=2cosh⁡z,e^z+e^{-z}=2\cosh z,

得到

p(y∣x1=s)∝exp⁡(sAσn2)cosh⁡(B+2sσn2).p(y\mid x_1=s) \propto \exp\left(\frac{sA}{\sigma_n^2}\right) \cosh\left(\frac{B+2s}{\sigma_n^2}\right).

所以決策規則為

x^1=+1\hat{x}_1=+1

若且唯若

2Aσn2+ln⁡cosh⁡(B+2σn2)cosh⁡(B−2σn2)>0;\frac{2A}{\sigma_n^2} + \ln \frac{ \cosh\left(\frac{B+2}{\sigma_n^2}\right) }{ \cosh\left(\frac{B-2}{\sigma_n^2}\right) } >0;

否則判定 x^1=−1\hat{x}_1=-1。

其中

A=y1−y2+y3,B=y2−y3+y4.A=y_1-y_2+y_3, \qquad B=y_2-y_3+y_4.

(c) 小 σn2\sigma_n^2 下的 ML 錯誤率近似

解題方法

當雜訊很小時,ML 決策主要由不同 x1x_1 類別之間距離最近的訊號向量決定。

四個可能的無雜訊接收向量如下:

(x1,x2)Hx(+1,+1)[1,0,0,1]T(+1,−1)[1,−2,2,−1]T(−1,+1)[−1,2,−2,1]T(−1,−1)[−1,0,0,−1]T\begin{array}{c|c} (x_1,x_2)&Hx\\ \hline (+1,+1)&[1,0,0,1]^T\\ (+1,-1)&[1,-2,2,-1]^T\\ (-1,+1)&[-1,2,-2,1]^T\\ (-1,-1)&[-1,0,0,-1]^T \end{array}

對於 AWGN,兩個訊號向量距離為 dd 時,二元近鄰錯誤率為

Q(d2σn).Q\left(\frac{d}{2\sigma_n}\right).

不同 x1x_1 類別的最近距離為:

  • (+1,+1)(+1,+1) 與 (−1,−1)(-1,-1):
d=22,d=2\sqrt{2},

因此錯誤率近似為

Q(2σn).Q\left(\frac{\sqrt{2}}{\sigma_n}\right).
  • (+1,−1)(+1,-1) 與 (−1,−1)(-1,-1),以及 (−1,+1)(-1,+1) 與 (+1,+1)(+1,+1):
d=23,d=2\sqrt{3},

因此錯誤率近似為

Q(3σn).Q\left(\frac{\sqrt{3}}{\sigma_n}\right).

四種 xx 等機率出現,所以

Pe(x1)≈14[2Q(2σn)+2Q(3σn)]=12Q(2σn)+12Q(3σn).\begin{aligned} P_e^{(x_1)} &\approx \frac{1}{4} \left[ 2Q\left(\frac{\sqrt{2}}{\sigma_n}\right) + 2Q\left(\frac{\sqrt{3}}{\sigma_n}\right) \right]\\ &= \frac{1}{2} Q\left(\frac{\sqrt{2}}{\sigma_n}\right) + \frac{1}{2} Q\left(\frac{\sqrt{3}}{\sigma_n}\right). \end{aligned}

其中第一項是主導項,因為 2<3\sqrt{2}<\sqrt{3}。


(d) 最小均方誤差的線性左逆系統

解題方法

在無雜訊時,

y=Hx.y=Hx.

因為 HH 具有滿 column rank,線性左逆可取 Moore–Penrose 偽逆:

W=(HTH)−1HT.W=(H^TH)^{-1}H^T.

先計算

HTH=[3−2−23],H^TH= \begin{bmatrix} 3&-2\\ -2&3 \end{bmatrix},

因此

(HTH)−1=15[3223].(H^TH)^{-1} = \frac{1}{5} \begin{bmatrix} 3&2\\ 2&3 \end{bmatrix}.

故

🔒

後續完整解題步驟與【答案】

免費註冊,享三天全站完整詳解閱覽。

免費註冊

第 1 題25 分

  1. (25%) Consider the transmission of an analog message signal m(t)m(t) over a noiseless channel via an impulse train sampler, i.e., the transmitted signal is given by
    x(t)=∑n=−∞∞m(nT)δ(t–nT)x(t) = \sum_{n=-\infty}^{\infty} m(nT)\delta(t – nT)
    where T>0T > 0 denotes the sampling period and δ(t)\delta(t) is the Dirac delta function. Upon receiving x(t)x(t), a reconstruction filter with frequency response
H(f)={T,∣f∣≤12T0,otherwiseH(f) = \begin{cases} T, & |f| \le \frac{1}{2T} \\ 0, & \text{otherwise} \end{cases}

is used to recover the message signal; denoted by m~(t)\tilde{m}(t) the filter output.

(a) (10%) Derive an expression for m~(t)\tilde{m}(t) in terms of the samples {m(nT):n∈Z}\{m(nT) : n \in \mathbb{Z}\}.
(b) (5%) Assume that m(t)m(t) is bandlimited to B(Hz)B(\text{Hz}) with B<1/(2T)B < 1/(2T). Determine whether m(t)m(t) can be perfectly reconstructed. Explain your answer.
(c) (10%) If only the signal values at the sampling instants t=nTt = nT are of interest, that is, we require
m~(nT)=m(nT),∀n∈Z,\tilde{m}(nT) = m(nT), \quad \forall n \in \mathbb{Z},
is the bandlimitedness assumption on m(t)m(t) necessary under this requirement?

登入後即可作答並保存紀錄。

這一題的完整詳解

此題考驗訊號取樣與重建的基本理論,包含 Nyquist 準則的應用以及取樣訊號在時域與頻域的表示。

核心觀念:取樣定理、理想低通濾波器重建、頻譜週期性。

(a) 取樣訊號 x(t)x(t) 的傅立葉轉換為 X(f)=1T∑k=−∞∞M(f−kT)X(f) = \frac{1}{T} \sum_{k=-\infty}^{\infty} M(f - \frac{k}{T}),其中 M(f)M(f) 是原始訊號 m(t)m(t) 的傅立葉轉換。
重建濾波器的頻率響應為 H(f)H(f)。重建訊號 m~(t)\tilde{m}(t) 的傅立葉轉換為 M~(f)=X(f)H(f)\tilde{M}(f) = X(f)H(f)。
M~(f)=(1T∑k=−∞∞M(f−kT))H(f)\tilde{M}(f) = \left( \frac{1}{T} \sum_{k=-\infty}^{\infty} M(f - \frac{k}{T}) \right) H(f)
由於 H(f)=TH(f) = T for ∣f∣≤1/(2T)|f| \le 1/(2T) 且在其他地方為 0,
M~(f)=1TH(f)∑k=−∞∞M(f−kT)=∑k=−∞∞M(f−kT)⋅H(f)T\tilde{M}(f) = \frac{1}{T} H(f) \sum_{k=-\infty}^{\infty} M(f - \frac{k}{T}) = \sum_{k=-\infty}^{\infty} M(f - \frac{k}{T}) \cdot \frac{H(f)}{T}
在 ∣f∣≤1/(2T)|f| \le 1/(2T) 的頻率範圍內,H(f)/T=1H(f)/T = 1。
M~(f)=∑k=−∞∞M(f−kT)for ∣f∣≤12T\tilde{M}(f) = \sum_{k=-\infty}^{\infty} M(f - \frac{k}{T}) \quad \text{for } |f| \le \frac{1}{2T}
然而,為了得到以取樣值表示的 m~(t)\tilde{m}(t),我們需要利用濾波器的脈衝響應。理想低通濾波器 H(f)H(f) 的脈衝響應為 h(t)=Tsinc(t/T)h(t) = T \text{sinc}(t/T)。
重建訊號 m~(t)\tilde{m}(t) 是取樣訊號 x(t)x(t) 通過脈衝響應為 h(t)h(t) 的線性系統的輸出。
m~(t)=x(t)∗h(t)=(∑n=−∞∞m(nT)δ(t−nT))∗(Tsinc(t/T))\tilde{m}(t) = x(t) * h(t) = \left(\sum_{n=-\infty}^{\infty} m(nT)\delta(t - nT)\right) * (T \text{sinc}(t/T))
m~(t)=T∑n=−∞∞m(nT)sinc(t−nTT)\tilde{m}(t) = T \sum_{n=-\infty}^{\infty} m(nT) \text{sinc}\left(\frac{t-nT}{T}\right)
m~(t)=T∑n=−∞∞m(nT)sinc(tT−n)\tilde{m}(t) = T \sum_{n=-\infty}^{\infty} m(nT) \text{sinc}\left(\frac{t}{T} - n\right)
其中 sinc(x)=sin⁡(πx)πx\text{sinc}(x) = \frac{\sin(\pi x)}{\pi x}。
【答案】m~(t)=T∑n=−∞∞m(nT)sinc(tT−n)\tilde{m}(t) = T \sum_{n=-\infty}^{\infty} m(nT) \text{sinc}\left(\frac{t}{T} - n\right)。

(b) 根據 Nyquist 取樣準則,若訊號 m(t)m(t) 的頻寬為 BB,則為了完美重建 m(t)m(t),取樣頻率 fs=1/Tf_s = 1/T 必須大於或等於 2B2B,即 B≤1/(2T)B \le 1/(2T)。
題目假設 m(t)m(t) 帶限於 B<1/(2T)B < 1/(2T)。
重建訊號的傅立葉轉換為 M~(f)=X(f)H(f)\tilde{M}(f) = X(f)H(f)。
X(f)=1T∑k=−∞∞M(f−kT)X(f) = \frac{1}{T} \sum_{k=-\infty}^{\infty} M(f - \frac{k}{T})
由於 B<1/(2T)B < 1/(2T),在 ∣f∣≤B|f| \le B 的頻率範圍內,原始訊號的頻譜 M(f)M(f) 與其頻移後的複製 M(f−k/T)M(f - k/T) (對於 k≠0k \neq 0) 不會重疊。
因此,在 ∣f∣≤B|f| \le B 的範圍內,X(f)=1TM(f)X(f) = \frac{1}{T} M(f)。
重建濾波器 H(f)=TH(f) = T 在 ∣f∣≤1/(2T)|f| \le 1/(2T) 的範圍內,且 B<1/(2T)B < 1/(2T)。
所以,在 ∣f∣≤B|f| \le B 的範圍內,
M~(f)=X(f)H(f)=(1TM(f))⋅T=M(f)\tilde{M}(f) = X(f)H(f) = \left(\frac{1}{T} M(f)\right) \cdot T = M(f)
由於 M(f)M(f) 在 ∣f∣>B|f| > B 時為零,且 H(f)H(f) 在 ∣f∣>1/(2T)|f| > 1/(2T) 時為零,因此 M~(f)\tilde{M}(f) 在 ∣f∣>B|f| > B 時也為零。

🔒

後續完整解題步驟與【答案】

免費註冊,享三天全站完整詳解閱覽。

免費註冊

第 2 題25 分

  1. (25%) Consider a conventional waveform channel y(t)=x(t)+n(t)y(t) = x(t)+n(t), where the noise process n(t)n(t) is described by a sequence of random pulse train given by
    n(t)=∑i=−∞∞nip(t−iT+δ)n(t) = \sum_{i=-\infty}^{\infty} n_i p(t-iT + \delta)
    Here, the waveform p(t)p(t) is a rectangular pulse with height 1/T1/\sqrt{T} on [0,T][0, T] and 00 elsewhere, the parameter δ∈[0,T)\delta \in [0, T) is a random time offset uniformly distributed on [0,T)[0, T), and the weighting coefficients nin_i's are independent and identically distributed Gaussian random variables with zero-mean and variance σn2\sigma_n^2. Assume that δ\delta is independent of nin_i's.

(a) (8%) Is the noise process n(t)n(t) wide-sense stationary (WSS)? Explain your answer.
(b) (6%) Describe the limiting behavior of the autocorrelation function and power spectral density of n(t)n(t) as T→0T \rightarrow 0.
(c) (4%) Let w(t)w(t) be the output of the ideal low-pass filter of bandwidth B(Hz)B(\text{Hz}) and height 11 with input n(t)n(t). Find the output variance of w(t)w(t) in the limit as T→0T \rightarrow 0.
(d) (7%) As T→0T \rightarrow 0, the noise process n(t)n(t) becomes Gaussian since any finite number of samples of n(t)n(t) are jointly Gaussian. Discuss whether w(t)w(t) defined in part (c) is also a Gaussian process and whether it is WSS. Explain your answer.

登入後即可作答並保存紀錄。

這一題的完整詳解

核心觀念

本題探討數位通訊與隨機程序中的經典模型——隨機脈衝序列(Random Pulse Train / PAM Signal) 的統計特性分析。核心觀念與定理包含:

  1. 隨機時延與廣義平穩性(WSS):一般的脈衝序列具有週期平穩性(Cyclostationary)。引入均勻分佈於 [0,T)[0, T) 的隨機時延 δ\delta(Random Time Offset / Jitter)後,可將週期非平穩過程轉化為嚴格的廣義平穩過程(Wide-Sense Stationary, WSS)。
  2. 全期望公式與區間拼接(Conditioning and Variable Substitution):計算隨機程序的平均值 E[n(t)]E[n(t)] 與自相關函數 Rn(t,t+τ)=E[n(t)n(t+τ)]R_n(t, t+\tau) = E[n(t)n(t+\tau)],透過條件期望值 E[⋅∣δ]E[\cdot \mid \delta] 與變數變換 θ=t−iT+δ\theta = t - iT + \delta,將離散級數和拼接為全實數軸連續積分。
  3. 白雜訊(White Noise)極限行為:當脈衝寬度 T→0T \to 0 且能量正規化時,自相關函數 Rn(τ)R_n(\tau) 趨近於狄拉克函數(Dirac Delta Function)δ(τ)\delta(\tau),其功率譜密度(Power Spectral Density, PSD)Sn(f)S_n(f) 趨近於全頻帶均勻分佈之常數。
  4. 線性時不變(LTI)系統之輸出特性:
    • 功率譜密度關係:Sw(f)=Sn(f)∣H(f)∣2S_w(f) = S_n(f)|H(f)|^2
    • 輸出方差(Variance):Var⁡(w(t))=Rw(0)=∫−∞∞Sw(f) df\operatorname{Var}(w(t)) = R_w(0) = \int_{-\infty}^{\infty} S_w(f) \, df
  5. 高斯程序(Gaussian Process)之線性不變性:高斯隨機程序通過任意穩定 LTI 系統後,其輸出仍為高斯隨機程序;若輸入為 WSS,則輸出亦必為 WSS。

詳細解題步驟與推導

(a) 判斷 n(t)n(t) 是否為廣義平穩過程(WSS)

Step 1:檢驗平均值 E[n(t)]E[n(t)]
由於 nin_i 與 δ\delta 相互獨立,且 E[ni]=0E[n_i] = 0,根據全期望公式:
E[n(t)]=Eδ[∑i=−∞∞E[ni]⋅p(t−iT+δ)]=Eδ[∑i=−∞∞0⋅p(t−iT+δ)]=0E[n(t)] = E_{\delta}\left[ \sum_{i=-\infty}^{\infty} E[n_i] \cdot p(t - iT + \delta) \right] = E_{\delta}\left[ \sum_{i=-\infty}^{\infty} 0 \cdot p(t - iT + \delta) \right] = 0
平均值為與時間 tt 無關的常數 00。

Step 2:計算自相關函數 Rn(t,t+τ)=E[n(t)n(t+τ)]R_n(t, t+\tau) = E[n(t)n(t+\tau)]
給定隨機變數 δ\delta 的條件下,由於 nin_i 彼此獨立且同分佈(i.i.d.),其二階互相關滿足 E[ninj]=σn2δijE[n_i n_j] = \sigma_n^2 \delta_{ij}(其中 δij\delta_{ij} 為 Kronecker delta):
E[n(t)n(t+τ)∣δ]=∑i=−∞∞∑j=−∞∞E[ninj]⋅p(t−iT+δ)p(t+τ−jT+δ)E[n(t)n(t+\tau) \mid \delta] = \sum_{i=-\infty}^{\infty} \sum_{j=-\infty}^{\infty} E[n_i n_j] \cdot p(t - iT + \delta) p(t + \tau - jT + \delta)
=σn2∑i=−∞∞p(t−iT+δ)p(t+τ−iT+δ)= \sigma_n^2 \sum_{i=-\infty}^{\infty} p(t - iT + \delta) p(t + \tau - iT + \delta)

對 δ∼Uniform[0,T)\delta \sim \text{Uniform}[0, T) 取期望值(其機率密度函數為 fδ(δ)=1T, δ∈[0,T)f_\delta(\delta) = \frac{1}{T}, \, \delta \in [0, T)):
Rn(t,t+τ)=σn2T∫0T∑i=−∞∞p(t−iT+δ)p(t+τ−iT+δ) dδR_n(t, t+\tau) = \frac{\sigma_n^2}{T} \int_{0}^{T} \sum_{i=-\infty}^{\infty} p(t - iT + \delta) p(t + \tau - iT + \delta) \, d\delta

令變數變換 θ=t−iT+δ\theta = t - iT + \delta,則 dθ=dδd\theta = d\delta。當 δ\delta 從 00 積分到 TT 時,θ\theta 的積分區間為 [t−iT, t−(i−1)T)[t - iT, \, t - (i-1)T)。將所有 i∈Zi \in \mathbb{Z} 的區間無縫拼接後,覆蓋整個實數軸 (−∞,∞)(-\infty, \infty):
Rn(t,t+τ)=σn2T∑i=−∞∞∫t−iTt−(i−1)Tp(θ)p(θ+τ) dθ=σn2T∫−∞∞p(θ)p(θ+τ) dθR_n(t, t+\tau) = \frac{\sigma_n^2}{T} \sum_{i=-\infty}^{\infty} \int_{t-iT}^{t-(i-1)T} p(\theta) p(\theta + \tau) \, d\theta = \frac{\sigma_n^2}{T} \int_{-\infty}^{\infty} p(\theta) p(\theta + \tau) \, d\theta

定義脈衝 p(t)p(t) 自身的自相關函數為 Rp(τ)=∫−∞∞p(θ)p(θ+τ) dθR_p(\tau) = \int_{-\infty}^{\infty} p(\theta) p(\theta + \tau) \, d\theta。
已知 p(t)=1Trect(t−T/2T)p(t) = \frac{1}{\sqrt{T}} \text{rect}\left(\frac{t - T/2}{T}\right),為寬度 TT、高度 1/T1/\sqrt{T} 之矩形脈衝,其自身卷積與積分結果為對稱三角脈衝:
Rp(τ)={1−∣τ∣T,∣τ∣≤T0,∣τ∣>T=tri⁡(τT)R_p(\tau) = \begin{cases} 1 - \frac{|\tau|}{T}, & |\tau| \le T \\ 0, & |\tau| > T \end{cases} = \operatorname{tri}\left(\frac{\tau}{T}\right)

代入得自相關函數為:
Rn(τ)=σn2Ttri⁡(τT)={σn2T(1−∣τ∣T),∣τ∣≤T0,∣τ∣>TR_n(\tau) = \frac{\sigma_n^2}{T} \operatorname{tri}\left(\frac{\tau}{T}\right) = \begin{cases} \frac{\sigma_n^2}{T} \left(1 - \frac{|\tau|}{T}\right), & |\tau| \le T \\ 0, & |\tau| > T \end{cases}

結論:
因為 E[n(t)]=0E[n(t)] = 0(常數)且 Rn(t,t+τ)=Rn(τ)R_n(t, t+\tau) = R_n(\tau) 僅取決於時間差 τ\tau,與絕對時間 tt 無關,故雜訊程序 n(t)n(t) 是廣義平穩程序(WSS)。


(b) 當 T→0T \rightarrow 0 時,Rn(τ)R_n(\tau) 與 Sn(f)S_n(f) 的極限行為

1. 自相關函數 Rn(τ)R_n(\tau) 的極限行為:
自相關函數為 Rn(τ)=σn2⋅[1Ttri⁡(τT)]R_n(\tau) = \sigma_n^2 \cdot \left[ \frac{1}{T} \operatorname{tri}\left(\frac{\tau}{T}\right) \right]。
函數 gT(τ)=1Ttri⁡(τT)g_T(\tau) = \frac{1}{T} \operatorname{tri}\left(\frac{\tau}{T}\right) 具備以下性質:

  • 寬度為 2T→T→002T \xrightarrow{T \to 0} 0
  • 峰值高度為 1T→T→0∞\frac{1}{T} \xrightarrow{T \to 0} \infty
  • 總面積為 ∫−∞∞gT(τ) dτ=1T⋅(12×2T×1)=1\int_{-\infty}^{\infty} g_T(\tau) \, d\tau = \frac{1}{T} \cdot \left(\frac{1}{2} \times 2T \times 1\right) = 1

依據廣義函數(Delta Sequences)定義,當 T→0T \to 0 時,gT(τ)→δ(τ)g_T(\tau) \to \delta(\tau)。
因此:

🔒

後續完整解題步驟與【答案】

免費註冊,享三天全站完整詳解閱覽。

免費註冊

第 3 題20 分

  1. (20%) Consider that a single-tone message signal m(t)=Amcos⁡(2πfmt)m(t) = A_m \cos(2\pi f_m t) is used to generate the following modulated signal with carrier amplitude AcA_c and frequency fc(≫fm)f_c (\gg f_m):
    s(t)=12AmAccos⁡[2π(fc+fm)t]+12(1−Am)Accos⁡[2π(fc−fm)t]s(t) = \frac{1}{2}A_m A_c \cos[2\pi(f_c+f_m)t] + \frac{1}{2}(1-A_m)A_c \cos[2\pi(f_c-f_m)t]
    where Am∈[0,1]A_m \in [0, 1] denotes the attenuation factor applied to the upper side frequency.

(a) (4%) Find the in-phase and quadrature components of s(t)s(t).
(b) (4%) Identify the modulation type represented by s(t)s(t) for Am=0A_m = 0, Am=0.5A_m = 0.5, Am=1A_m = 1, and for other values of aa. (Note: the problem statement uses AmA_m for attenuation factor, but asks for values of aa. Assuming aa refers to AmA_m).
(c) (8%) Assume that the signal s(t)s(t) is, plus the carrier Accos⁡(2πfct)A_c \cos(2\pi f_c t), is sent and passed through an envelope detector. In the absence of channel noise, determine the envelope distortion introduced by the quadrature component in the detector output.
(d) (4%) Based on your result in part (c), determine the values of AmA_m for which the distortion is eliminated and the values for which it is maximized. Justify your answer.

登入後即可作答並保存紀錄。

這一題的完整詳解

核心觀念

本題考查:

  1. 利用和差公式將上下邊頻分解成同相(in-phase)與正交(quadrature)分量。
  2. 由上下邊頻的相對振幅判斷調變種類。
  3. 包絡檢波器實際量測的是射頻訊號的包絡:
    e(t)=xI2(t)+xQ2(t)e(t)=\sqrt{x_I^2(t)+x_Q^2(t)}
  4. 正交分量無法直接以線性方式通過包絡檢波器,會以平方項形式造成包絡失真。

令

ωc=2πfc,ωm=2πfm\omega_c=2\pi f_c,\qquad \omega_m=2\pi f_m

並將 attenuation factor 統一記為 AmA_m。


(a)同相與正交分量

原訊號為

s(t)=12AmAccos⁡[(ωc+ωm)t]+12(1−Am)Accos⁡[(ωc−ωm)t].s(t)=\frac{1}{2}A_mA_c\cos[(\omega_c+\omega_m)t] +\frac{1}{2}(1-A_m)A_c\cos[(\omega_c-\omega_m)t].

利用

cos⁡(ωc+ωm)t=cos⁡ωctcos⁡ωmt−sin⁡ωctsin⁡ωmt\cos(\omega_c+\omega_m)t =\cos\omega_ct\cos\omega_mt-\sin\omega_ct\sin\omega_mt

以及

cos⁡(ωc−ωm)t=cos⁡ωctcos⁡ωmt+sin⁡ωctsin⁡ωmt,\cos(\omega_c-\omega_m)t =\cos\omega_ct\cos\omega_mt+\sin\omega_ct\sin\omega_mt,

可得

s(t)=Ac2Am(cos⁡ωctcos⁡ωmt−sin⁡ωctsin⁡ωmt)+Ac2(1−Am)(cos⁡ωctcos⁡ωmt+sin⁡ωctsin⁡ωmt).\begin{aligned} s(t) ={}&\frac{A_c}{2}A_m \left(\cos\omega_ct\cos\omega_mt-\sin\omega_ct\sin\omega_mt\right)\\ &+\frac{A_c}{2}(1-A_m) \left(\cos\omega_ct\cos\omega_mt+\sin\omega_ct\sin\omega_mt\right). \end{aligned}

整理 cos⁡ωct\cos\omega_ct 與 sin⁡ωct\sin\omega_ct 的係數:

s(t)=Ac2cos⁡ωmtcos⁡ωct+Ac2(1−2Am)sin⁡ωmtsin⁡ωct.s(t)=\frac{A_c}{2}\cos\omega_mt\cos\omega_ct +\frac{A_c}{2}(1-2A_m)\sin\omega_mt\sin\omega_ct.

採用通訊系統常見表示式

s(t)=Ac[I(t)cos⁡ωct−Q(t)sin⁡ωct],s(t)=A_c\left[I(t)\cos\omega_ct-Q(t)\sin\omega_ct\right],

因此

I(t)=12cos⁡ωmtI(t)=\frac{1}{2}\cos\omega_mt

且

Q(t)=2Am−12sin⁡ωmt.Q(t)=\frac{2A_m-1}{2}\sin\omega_mt.

所以:

  • 同相分量:
I(t)=12cos⁡(2πfmt)\boxed{I(t)=\frac{1}{2}\cos(2\pi f_mt)}
  • 正交分量:
Q(t)=2Am−12sin⁡(2πfmt)\boxed{Q(t)=\frac{2A_m-1}{2}\sin(2\pi f_mt)}

若採用 s(t)=Ac[I(t)cos⁡ωct+Q(t)sin⁡ωct]s(t)=A_c[I(t)\cos\omega_ct+Q(t)\sin\omega_ct] 的符號慣例,正交分量會寫成 1−2Am2sin⁡ωmt\frac{1-2A_m}{2}\sin\omega_mt;物理結果完全相同。


(b)調變種類判斷

上下邊頻的振幅分別為

AUSB=12AmAcA_{\mathrm{USB}}=\frac{1}{2}A_mA_c

以及

ALSB=12(1−Am)Ac.A_{\mathrm{LSB}}=\frac{1}{2}(1-A_m)A_c.

題目中的訊號沒有獨立載波,因此屬於 suppressed-carrier 類型。

Am=0A_m=0

此時

AUSB=0,ALSB=12Ac.A_{\mathrm{USB}}=0,\qquad A_{\mathrm{LSB}}=\frac{1}{2}A_c.

只剩下低側頻,因此為:

下邊帶抑制載波調變(LSB-SC)\boxed{\text{下邊帶抑制載波調變(LSB-SC)}}

Am=0.5A_m=0.5

此時

AUSB=ALSB=14Ac.A_{\mathrm{USB}}=A_{\mathrm{LSB}}=\frac{1}{4}A_c.

上下邊頻振幅相等,且沒有載波,因此為:

雙邊帶抑制載波調變(DSB-SC)\boxed{\text{雙邊帶抑制載波調變(DSB-SC)}}

此時由(a)可見 Q(t)=0Q(t)=0,訊號完全是同相分量。

Am=1A_m=1

此時

AUSB=12Ac,ALSB=0.A_{\mathrm{USB}}=\frac{1}{2}A_c,\qquad A_{\mathrm{LSB}}=0.

只剩下高側頻,因此為:

上邊帶抑制載波調變(USB-SC)\boxed{\text{上邊帶抑制載波調變(USB-SC)}}

其他情況

當

0<Am<1,Am≠0.5,0<A_m<1,\qquad A_m\neq 0.5,

上下邊頻同時存在,但振幅不相等,故為:

非對稱雙邊帶抑制載波調變\boxed{\text{非對稱雙邊帶抑制載波調變}}

也可稱為 asymmetric DSB-SC。


(c)包絡檢波器造成的失真

題目指定將載波

Accos⁡ωctA_c\cos\omega_ct

與 s(t)s(t) 相加,因此總訊號為

x(t)=Accos⁡ωct+s(t).x(t)=A_c\cos\omega_ct+s(t).

代入(a)的結果:

🔒

後續完整解題步驟與【答案】

免費註冊,享三天全站完整詳解閱覽。

免費註冊

第 4 題30 分

  1. (30%) Consider the discrete-time channel
    y=h∗x+ny = h * x + n
    where "*" is the convolution operator, h=[1,−1,1]h = [1, -1, 1] denotes the channel impulse response, x=[x1,x2]x = [x_1, x_2] represents the transmitted vector with two independent and equiprobable signals selected from the set {+1,−1}\{+1, -1\}, and the noise vector nn consists of independent and identically distributed Gaussian random variables with zero-mean and variance σn2\sigma_n^2. Assume full linear convolution with zero padding. Answer the following questions:

(a) (2%) Express yy in matrix form y=Hx+ny = Hx+n by specifying the channel matrix HH.
(b) (5%) Based on the representation in part (a), derive the maximum-likelihood (ML) decision rule for estimating x1x_1 from yy, given HH and the noise statistics.
(c) (5%) Assume that σn2\sigma_n^2 is sufficiently small, derive an approximation of the ML decoding error probability for x1x_1 under the decision rule obtained in part (b). Express your result in terms of the Q-function, where Q(t)=∫t∞12πe−u2/2duQ(t) = \int_t^\infty \frac{1}{\sqrt{2\pi}} e^{-u^2/2} du.
(d) (5%) In the absence of noise, design a linear left-inverse system WW that minimizes the mean-squared error Ex[∥Wy−x∣∣2]E_x [\| Wy - x ||^2], where the expectation is taken over all equiprobable transmitted vectors xx and ∣∣⋅∣∣|| \cdot || denotes the Euclidean norm.
(e) (5%) Let y~=Wy\tilde{y} = Wy denote the output vector of the inverse system in part (d). Determine the probability distribution of the effective noise vector n~=Wn\tilde{n} = Wn.
(f) (5%) Suppose that the optimal decoding of x1x_1 is derived based on the scalar observation y1=x1+n~1y_1 = x_1 + \tilde{n}_1. What is the decoding error probability for sufficiently small σn2\sigma_n^2? Express your result in terms of the Q-function defined in part (c).
(g) (3%) Even though both decoders are optimal under their respective models, one based on yy and the other based on y~\tilde{y}, their performance is not the same. Explain why.

登入後即可作答並保存紀錄。

這一題的完整詳解

核心觀念

本題綜合考查:

  • 線性卷積的矩陣表示。
  • AWGN 通道下的最大概似決策。
  • 高訊噪比下,以訊號集合間最小歐氏距離近似錯誤率。
  • Moore–Penrose 線性左逆矩陣。
  • 高斯雜訊經線性轉換後的分布。
  • 雜訊增強與充分統計量的概念。

令

h=[1,−1,1],x=[x1,x2]T.h=[1,-1,1],\qquad x=[x_1,x_2]^T.

採用完整線性卷積並補零,因此輸出長度為 3+2−1=43+2-1=4。


(a) 矩陣形式

完整卷積為

y1=x1+n1,y2=−x1+x2+n2,y3=x1−x2+n3,y4=x2+n4.\begin{aligned} y_1&=x_1+n_1,\\ y_2&=-x_1+x_2+n_2,\\ y_3&=x_1-x_2+n_3,\\ y_4&=x_2+n_4. \end{aligned}

因此

y=Hx+n,y=Hx+n,

其中

H=[10−111−101].H= \begin{bmatrix} 1&0\\ -1&1\\ 1&-1\\ 0&1 \end{bmatrix}.

令 HH 的兩個 column vector 為

a=[1−110],b=[01−11],a= \begin{bmatrix} 1\\-1\\1\\0 \end{bmatrix}, \qquad b= \begin{bmatrix} 0\\1\\-1\\1 \end{bmatrix},

則

y=ax1+bx2+n.y=ax_1+bx_2+n.

(b) x1x_1 的最大概似決策規則

解題方法

因為 x2x_2 是未知干擾訊號,估計 x1x_1 時必須將 x2x_2 邊際化。由於 x1,x2x_1,x_2 獨立且等機率,最大概似規則為

x^1=arg⁡max⁡s∈{+1,−1}p(y∣x1=s),\hat{x}_1 = \arg\max_{s\in\{+1,-1\}}p(y\mid x_1=s),

其中

p(y∣x1=s)=12∑t∈{+1,−1}p(y∣x1=s,x2=t).p(y\mid x_1=s) = \frac{1}{2} \sum_{t\in\{+1,-1\}} p(y\mid x_1=s,x_2=t).

AWGN 的條件機率密度為

p(y∣x1=s,x2=t)∝exp⁡(−∥y−sa−tb∥22σn2).p(y\mid x_1=s,x_2=t) \propto \exp\left( -\frac{\|y-sa-tb\|^2}{2\sigma_n^2} \right).

因此可寫成

x^1=arg⁡max⁡s∈{+1,−1}∑t=±1exp⁡(−∥y−sa−tb∥22σn2).\hat{x}_1 = \arg\max_{s\in\{+1,-1\}} \sum_{t=\pm1} \exp\left( -\frac{\|y-sa-tb\|^2}{2\sigma_n^2} \right).

計算所需內積:

aTa=3,bTb=3,aTb=−2.a^Ta=3,\qquad b^Tb=3,\qquad a^Tb=-2.

令

A=aTy,B=bTy.A=a^Ty,\qquad B=b^Ty.

則

∥y−sa−tb∥2=∥y∥2+6+2st−2sA−2tB.\|y-sa-tb\|^2 = \|y\|^2+6+2st-2sA-2tB.

去除與 ss 無關的共同項後,

p(y∣x1=s)∝exp⁡(sAσn2)∑t=±1exp⁡(t(B+2s)σn2).p(y\mid x_1=s) \propto \exp\left(\frac{sA}{\sigma_n^2}\right) \sum_{t=\pm1} \exp\left(\frac{t(B+2s)}{\sigma_n^2}\right).

利用

ez+e−z=2cosh⁡z,e^z+e^{-z}=2\cosh z,

得到

p(y∣x1=s)∝exp⁡(sAσn2)cosh⁡(B+2sσn2).p(y\mid x_1=s) \propto \exp\left(\frac{sA}{\sigma_n^2}\right) \cosh\left(\frac{B+2s}{\sigma_n^2}\right).

所以決策規則為

x^1=+1\hat{x}_1=+1

若且唯若

2Aσn2+ln⁡cosh⁡(B+2σn2)cosh⁡(B−2σn2)>0;\frac{2A}{\sigma_n^2} + \ln \frac{ \cosh\left(\frac{B+2}{\sigma_n^2}\right) }{ \cosh\left(\frac{B-2}{\sigma_n^2}\right) } >0;

否則判定 x^1=−1\hat{x}_1=-1。

其中

A=y1−y2+y3,B=y2−y3+y4.A=y_1-y_2+y_3, \qquad B=y_2-y_3+y_4.

(c) 小 σn2\sigma_n^2 下的 ML 錯誤率近似

解題方法

當雜訊很小時,ML 決策主要由不同 x1x_1 類別之間距離最近的訊號向量決定。

四個可能的無雜訊接收向量如下:

(x1,x2)Hx(+1,+1)[1,0,0,1]T(+1,−1)[1,−2,2,−1]T(−1,+1)[−1,2,−2,1]T(−1,−1)[−1,0,0,−1]T\begin{array}{c|c} (x_1,x_2)&Hx\\ \hline (+1,+1)&[1,0,0,1]^T\\ (+1,-1)&[1,-2,2,-1]^T\\ (-1,+1)&[-1,2,-2,1]^T\\ (-1,-1)&[-1,0,0,-1]^T \end{array}

對於 AWGN,兩個訊號向量距離為 dd 時,二元近鄰錯誤率為

Q(d2σn).Q\left(\frac{d}{2\sigma_n}\right).

不同 x1x_1 類別的最近距離為:

  • (+1,+1)(+1,+1) 與 (−1,−1)(-1,-1):
d=22,d=2\sqrt{2},

因此錯誤率近似為

Q(2σn).Q\left(\frac{\sqrt{2}}{\sigma_n}\right).
  • (+1,−1)(+1,-1) 與 (−1,−1)(-1,-1),以及 (−1,+1)(-1,+1) 與 (+1,+1)(+1,+1):
d=23,d=2\sqrt{3},

因此錯誤率近似為

Q(3σn).Q\left(\frac{\sqrt{3}}{\sigma_n}\right).

四種 xx 等機率出現,所以

Pe(x1)≈14[2Q(2σn)+2Q(3σn)]=12Q(2σn)+12Q(3σn).\begin{aligned} P_e^{(x_1)} &\approx \frac{1}{4} \left[ 2Q\left(\frac{\sqrt{2}}{\sigma_n}\right) + 2Q\left(\frac{\sqrt{3}}{\sigma_n}\right) \right]\\ &= \frac{1}{2} Q\left(\frac{\sqrt{2}}{\sigma_n}\right) + \frac{1}{2} Q\left(\frac{\sqrt{3}}{\sigma_n}\right). \end{aligned}

其中第一項是主導項,因為 2<3\sqrt{2}<\sqrt{3}。


(d) 最小均方誤差的線性左逆系統

解題方法

在無雜訊時,

y=Hx.y=Hx.

因為 HH 具有滿 column rank,線性左逆可取 Moore–Penrose 偽逆:

W=(HTH)−1HT.W=(H^TH)^{-1}H^T.

先計算

HTH=[3−2−23],H^TH= \begin{bmatrix} 3&-2\\ -2&3 \end{bmatrix},

因此

(HTH)−1=15[3223].(H^TH)^{-1} = \frac{1}{5} \begin{bmatrix} 3&2\\ 2&3 \end{bmatrix}.

故

🔒

後續完整解題步驟與【答案】

免費註冊,享三天全站完整詳解閱覽。

免費註冊

其他考古題

115 年中正大學的其他科目

其他學校的通訊系考古題