Digital Signal Processing Notes

1. Introduction to Digital Signals and Systems

• Analog Signal: Continuous in time and amplitude.

• Digital Signal: Discrete in time (sampled) and amplitude (quantized).

• Advantages of DSP: Flexibility, accuracy, perfect reproducibility,

size/cost, advanced algorithms.

• Disadvantages of DSP: Bandwidth limitation, quantization effects,

complex hardware/software.

• DSP System Components:

◦ Analog-to-Digital Converter (ADC): Sampling, Quantization, Encoding.

◦ Digital Signal Processor (Microprocessor/FPGA/ASIC).

◦ Digital-to-Analog Converter (DAC): Reconstruction.

◦ Anti-aliasing filter (before ADC), Reconstruction filter (after DAC).

2. Discrete-Time Signals and Systems

• Discrete-Time (DT) Signal: x[n] defined only for integer values of n.

• Basic DT Signals:

◦ Unit Impulse: δ[n] = {1 for n=0, 0 otherwise}.

◦ Unit Step: u[n] = {1 for n≥0, 0 otherwise}.

◦ Real Exponential: x[n] = aⁿ.

◦ Complex Exponential: x[n] = A e^(jω₀n + φ).

• DT System Properties:

◦ Linearity: T{ax₁[n] + bx₂[n]} = aT{x₁[n]} + bT{x₂[n]}.

◦ Time Invariance: T{x[n-n₀]} = y[n-n₀].

◦ Causality: Output depends only on present and past inputs.

◦ Stability (BIBO): Bounded input produces bounded output.

◦ Memory/Memoryless: Output depends on past/future inputs or only
present input.

• Linear Time-Invariant (LTI) Systems:

◦ Characterized by Impulse Response h[n].

◦ Convolution Sum: y[n] = x[n] * h[n] = Σ (k=-∞ to ∞) x[k]h[n-k].

◦ Causality: h[n] = 0 for n < 0.

◦ Stability: Σ (n=-∞ to ∞) |h[n]| < ∞.

• Finite Impulse Response (FIR) Systems: Impulse response h[n] has

finite duration.

◦ Always stable. Can be designed to have linear phase.

• Infinite Impulse Response (IIR) Systems: Impulse response h[n] has

infinite duration.

◦ More efficient in terms of components/memory for given magnitude

response.

◦ Can be unstable. Cannot have perfect linear phase.

• Difference Equations: Describe DT LTI systems.

◦ y[n] = -Σ (k=1 to N) a_k y[n-k] + Σ (k=0 to M) b_k x[n-k].

3. Sampling and Reconstruction

• Sampling: Converting a continuous-time signal x_c(t) into a discrete-time

signal x[n].

◦ x[n] = x_c(nT_s) where T_s is sampling period, f_s = 1/T_s is sampling

frequency.

• Nyquist-Shannon Sampling Theorem: A continuous-time signal x_c(t)

with maximum frequency f_max can be perfectly reconstructed from its
samples x[n] if f_s > 2f_max (or ω_s > 2ω_max).

◦ Nyquist Rate: 2f_max.

• Aliasing: If f_s ≤ 2f_max, higher frequency components in x_c(t) will

appear as lower frequency components in x[n], leading to irreversible
distortion.

• Anti-Aliasing Filter: An analog low-pass filter applied before sampling to

limit the bandwidth of x_c(t) and prevent aliasing.
 • Reconstruction: Converting a discrete-time signal x[n] back to a
continuous-time signal x_r(t).

◦ Ideally, done using an ideal low-pass filter (interpolation) with cutoff

frequency ω_c such that ω_max ≤ ω_c ≤ ω_s - ω_max.

◦ Practical DACs use zero-order hold or first-order hold.

4. Z-Transform

• Definition: X(z) = Σ (n=-∞ to ∞) x[n] z^(-n).

◦ z is a complex variable.

• Region of Convergence (ROC): Range of 'z' for which the sum

converges.

◦ For causal signals, ROC is exterior to a circle.

◦ For stable systems, ROC includes the unit circle (|z|=1).

• Inverse Z-Transform: Partial fraction expansion, power series

expansion, contour integration.

• Properties: Linearity, Time Shift, Scaling, Differentiation, Convolution

(x[n]*h[n] <=> X(z)H(z)).

• System Function H(z): For LTI systems, H(z) = Y(z)/X(z).

◦ Poles: Roots of the denominator of H(z).

◦ Zeros: Roots of the numerator of H(z).

◦ Stability: A causal LTI system is stable if and only if all its poles lie strictly
inside the unit circle in the z-plane.

• Frequency Response: H(e^(jω)) obtained by evaluating H(z) on the unit

circle (z = e^(jω)).

5. Discrete Fourier Transform (DFT) and Fast Fourier

Transform (FFT)

• Discrete Fourier Transform (DFT): Transforms a finite-length discrete-

time sequence x[n] (of length N) into a finite-length frequency-domain
sequence X[k] (of length N).

◦ X[k] = Σ (n=0 to N-1) x[n] e^(-j2πkn/N) for k = 0, 1, ..., N-1.

◦ Inverse DFT (IDFT): x[n] = (1/N) Σ (k=0 to N-1) X[k] e^(j2πkn/N) for n =
0, 1, ..., N-1.

• Properties of DFT: Linearity, Circular Shift, Circular Convolution.

• Circular Convolution: Convolution in the time domain corresponds to

multiplication in the DFT domain, but it's circular convolution.

◦ To get linear convolution from DFT, use zero-padding to make lengths

M+N-1.

• Fast Fourier Transform (FFT): An efficient algorithm to compute the

DFT.

◦ Reduces computational complexity from O(N²) to O(N log₂N).

◦ Common algorithms: Radix-2 (decimation-in-time, decimation-in-

frequency).

• Applications of DFT/FFT: Spectral analysis, filtering, correlation, fast

convolution.

6. Digital Filter Design

• Ideal Filter Characteristics: Perfect passband, perfect stopband, sharp

transition band, linear phase. Non-causal, non-realizable.

• FIR Filter Design:

◦ Windowing Method: Truncates an ideal (infinite) impulse response by

multiplying with a finite-length window function (e.g., Rectangular,
Hanning, Hamming, Blackman).

Trade-offs: Main lobe width vs. side lobe magnitude.

Always stable, can achieve linear phase.

◦ Frequency Sampling Method: Specifies desired frequency response at

certain points and uses IDFT to find impulse response.

◦ Optimal (Parks-McClellan/Remez) Algorithm: Designs filters that

minimize the maximum error between desired and actual response
(equiripple design).

• IIR Filter Design:

◦ Typically designed from analog prototypes using transformations.

◦ Impulse Invariance: h[n] = h_c(nT_s). Preserves impulse response.

Suffers from aliasing if analog filter is not band-limited.

◦ Bilinear Transformation: s = (2/T_s) * (1-z⁻¹)/(1+z⁻¹). Maps the entire
left-half s-plane to the unit circle.

Avoids aliasing but introduces frequency warping (non-linear mapping

of analog to digital frequencies).

Must pre-warp analog filter specifications.

◦ Analog Filter Prototypes: Butterworth, Chebyshev (Type I & II), Elliptic.

Butterworth: Maximally flat response in passband, monotonic rolloff.

Chebyshev: Equiripple in passband (Type I) or stopband (Type II),

steeper rolloff.

Elliptic: Equiripple in both passband and stopband, steepest rolloff for

given order.

7. Filter Structures

• Direct Form I & II: Basic implementations from difference equations.

◦ Direct Form II is canonical (uses fewer delay elements).

• Cascade Form: Connecting lower-order sections in series.

◦ Used for higher-order filters, can improve numerical stability.

• Parallel Form: Connecting lower-order sections in parallel (for IIR filters,

using partial fraction expansion).

◦ Used for IIR filters, can improve numerical stability.

• Lattice Structures: Specialized structures, particularly useful in adaptive

filtering and speech processing.

8. Multirate Digital Signal Processing

• Decimation (Downsampling): Reducing the sampling rate of a DT

signal.

◦ x_d[m] = x[mM] where M is the decimation factor (M > 1).

◦ Requires an anti-aliasing (decimation) low-pass filter before downsampling

to prevent aliasing.

• Interpolation (Upsampling): Increasing the sampling rate of a DT

signal.
◦ x_e[n] = {x[n/L] if n is a multiple of L, 0 otherwise}. L is the interpolation
factor (L > 1).

◦ Followed by an interpolation (reconstruction) low-pass filter to smooth the

signal.

• Applications: Sample rate conversion, subband coding, filter banks,

efficient implementation of narrow-band filters.

9. Quantization Effects

• Quantization: Converting continuous amplitude values to discrete levels.

• Quantization Error: Difference between the unquantized and quantized

signal.

• Quantization Noise: Modeled as a white noise source added to the

signal.

◦ Signal-to-Quantization-Noise Ratio (SQNR): For B bits, SQNR ≈ 6.02B +

1.76 dB.

• Coefficient Quantization: Quantizing filter coefficients affects filter

response.

• Product Quantization (Round-off Noise): Quantizing results of

multiplications in filter implementations.

• Overflow: Occurs when results of additions exceed the maximum

representable value.

◦ Saturation arithmetic can prevent limit cycles but introduces distortion.

• Limit Cycles: Undesirable oscillations in IIR filters due to non-linearities

from quantization or overflow.