Complete Formulas for Continuous and Discrete-Time
Sampling
Signal Processing Reference
Part 1: Continuous-Time (C-T) Sampling
1.1 The Sampling Process
Ideal Sampling Model:
∞
X
xs (t) = xc (t) · s(t) = xc (t) · δ(t − nT )
n=−∞
where:
• T : sampling period (seconds)
• fs = T1 : sampling frequency (Hz)
• Ωs = 2πfs = 2π
T
: sampling angular frequency (rad/s)
Sampled Signal (Discrete-Time Sequence):
x[n] = xc (nT )
1.2 Frequency Domain Analysis
Fourier Transform of the Impulse Train:
∞
2π X
S(jΩ) = δ(Ω − kΩs )
T k=−∞
Fourier Transform of the Sampled Signal xs (t):
∞
1 1 X
Xs (jΩ) = Xc (jΩ) ∗ S(jΩ) = Xc (j(Ω − kΩs ))
2π T k=−∞
Key Insight: The spectrum Xs (jΩ) consists of copies of the original spectrum
Xc (jΩ), scaled by 1/T , and shifted by integer multiples of Ωs .
1
1.3 The Nyquist-Shannon Sampling Theorem
If a continuous-time signal xc (t) is bandlimited (i.e., Xc (jΩ) = 0 for |Ω| ≥ ΩN ), then
xc (t) can be uniquely reconstructed from its samples x[n] = xc (nT ) if:
Ωs > 2ΩN or equivalently fs > 2fN
where:
• ΩN , fN : highest frequencies in the signal (Nyquist frequency)
• Nyquist Rate: 2fN
• Nyquist Frequency (Folding Frequency): fs /2
1.4 Aliasing
Cause: Violation of sampling theorem (fs ≤ 2fN )
Effect: Spectral copies overlap, causing high-frequency components to appear as
lower frequencies. This distortion is irreversible.
1.5 Ideal Reconstruction (Interpolation)
Ideal Interpolation Filter:
(
T, |Ω| ≤ Ωs /2
Hr (jΩ) =
0, |Ω| > Ωs /2
where ΩN < Ωc < Ωs − ΩN (commonly Ωc = Ωs /2).
Reconstructed Signal in Frequency Domain:
Xr (jΩ) = Xs (jΩ) · Hr (jΩ)
If no aliasing occurred, Xr (jΩ) = Xc (jΩ).
Reconstructed Signal in Time Domain (Sinc Interpolation):
∞
X π
xr (t) = x[n] · sinc (t − nT )
n=−∞
T
sin(u)
where sinc(u) = u
.
Part 2: Discrete-Time (D-T) Sampling (Decimation
& Interpolation)
2.1 The Downsampling Process
Definition:
xd [m] = x[mM ]
where M is the downsampling factor.
2
2.2 Frequency Domain Analysis
Intermediate signal with zeros:
(
x[n], if n/M is an integer
xp [n] = = c[n] · x[n]
0, otherwise
DTFT of xp [n]:
M −1
jω 1 X
Xp (e ) = X(ej(ω−2πk)/M )
M k=0
DTFT of the Downsampled Signal xd [m]:
M −1
jω jω/M 1 X
Xd (e ) = Xp (e )= X(ej(ω−2πk)/M )
M k=0
Key Insight: The spectrum Xd (ejω ) is an average of the original spectrum and its
M − 1 aliases.
2.3 Avoiding Aliasing in Discrete-Time
Condition for No Aliasing: If X(ejω ) = 0 for ωN ≤ |ω| ≤ π, then downsampling by
M will not cause aliasing if:
π
ωN <
M
Anti-Aliasing Filter:
(
1, |ω| ≤ π/M
Haa (ejω ) =
0, otherwise
2.4 Upsampling (Increasing the Sampling Rate)
Definition (by factor L):
(
x[n/L], if n/L is an integer
xu [n] =
0, otherwise
DTFT of the Upsampled Signal:
Xu (ejω ) = X(ejωL )
Anti-Imaging Filter:
(
L, |ω| ≤ π/L
Hint (ejω ) =
0, otherwise
3
Concept Continuous-Time Sampling Discrete-Time Sampling
P (Downsampling)
Process · δ(t − nT )
xs (t) = xc (t)P xd [m] = x[mM ]
Frequency Do- Xs (jΩ) = T1 k Xc (j(Ω − kΩs )) XdP (ejω ) =
1 M −1 j(ω−2πk)/M
main Result M k=0 X(e )
Aliasing Con- Occurs if fs ≤ 2fN Occurs if original bandwidth
dition ωN ≥ π/M
Anti-Aliasing Ωs > 2ΩN ωN < π/M
Condition
Reconstruction Ideal LPF: Hr (jΩ) with cut-off Upsampling + LPF: Hint (ejω )
Ωs /2 P with cut-off π/L
Interpolation xr (t) = x[n] · sinc( Tπ (t − nT )) Implemented via digital filter af-
Formula ter upsampling
Summary of Key Results
Important Relationships
Frequency Relationships
• Continuous-Time: Ω = 2πf , Ωs = 2πfs
• Discrete-Time: ω = ΩT = Ω
fs
• Normalized Frequency: ω = 2π ffs
Reconstruction Formulas
Ideal Reconstruction Filter Impulse Response:
πt sin(πt/T )
hr (t) = sinc =
T πt/T
Zero-Order Hold (ZOH) Reconstruction:
(
1, 0 ≤ t < T
hzoh (t) =
0, otherwise
Practical Considerations
Oversampling Ratio:
fs
OSR =
2fN
Anti-aliasing Filter Requirements:
• Stopband frequency: fstop ≤ fs − fN
• Passband frequency: fpass ≤ fN
• Transition band: fN < f < fs − fN