UNIT – 1

1. Sequences:

 A discrete-time signal is a sequence of numbers x[n], where n is an integer

index representing the discrete time instants.
 Examples: Speech samples, image pixels, sensor readings, financial data.
 Basic operations: Addition, subtraction, scaling, shifting, convolution.

2. Representation on Orthogonal Basis:

 Any discrete-time signal can be represented as a linear combination of basis

functions.
 Common basis functions: Fourier series (sinusoids), Walsh functions
(rectangular pulses), wavelet functions.
 Coefficients of the basis functions determine the frequency content of the
signal.

3. Sampling and Reconstruction:

 Sampling: Converting a continuous-time signal to a discrete-time signal by

taking values at regular intervals (T seconds apart).
 Nyquist-Shannon Sampling Theorem: Sampling rate must be greater than
twice the highest frequency present in the signal for perfect reconstruction.
 Reconstruction: Recovering the continuous-time signal from its samples using
filters and interpolation techniques.

4. Discrete Systems Attributes:

 Linearity: The output is a scaled and shifted version of the input.

 Time-invariance: The output for a delayed input is a delayed version of the
original output.
 Causality: The output only depends on present and past inputs.
  Stability: The output remains bounded for bounded inputs.

5. Z-Transform:

 A powerful tool for analyzing discrete-time systems.

 Maps a sequence x[n] to a function X(z) in the z-plane.
 Z-transform properties: Linearity, time-shifting, scaling, differentiation,
integration.
 Used for solving difference equations, analyzing frequency response, and
designing filters.

6. Analysis of LSI Systems:

 Linear Shift-Invariant (LSI) systems are the most common type of discrete-
time systems.
 Characterized by impulse response h[n], which defines the system's output
for any input.
 Convolution: Output of an LSI system for any input is the convolution of the
input and the impulse response.
 Frequency response: H(z) is the Z-transform of the impulse response,
providing information about the system's behavior at different frequencies.

7. Frequency Analysis:

 Fourier Transform: Decomposes a discrete-time signal into its frequency

components.
 Discrete Fourier Transform (DFT): Computes the Fourier coefficients of a
finite-length sequence.
 Fast Fourier Transform (FFT): Efficient algorithm for computing the DFT.
 Used for spectral analysis, filter design, and signal processing applications.

8. Inverse Systems:

 An inverse system undoes the operation of another system.

 Not always possible, depends on the system's properties.
 Applications: Error correction, communication systems, control systems.
 UNIT – 2

Discrete Fourier Transform (DFT): Unveiling the

Frequency Spectrum
1. Definition and Importance:

 The DFT transforms a finite sequence of N real or complex numbers (x[n])

into another sequence of N complex numbers (X[k]).
 These complex numbers represent the magnitude and phase of the frequency
components present in the original signal.
 Applications: Spectral analysis, filter design, image processing, data
compression, communication systems.

2. Formula and Key Points:

 DFT formula:
X[k] = ∑(n=0 to N-1) x[n] * W_N^-kn, where W_N = e^(-j2π/N)

 k represents the frequency index (0 to N-1).

 W_N is the Nth root of unity, a complex number used for efficient
computation.
 Properties: Linearity, periodicity, symmetry, convolution theorem.

3. Visualization and Interpretation:

 Magnitude of X[k] indicates the strength of the kth frequency component.

 Phase of X[k] represents the time delay of the kth frequency component.
 Plot of |X[k]| vs. k is called the magnitude spectrum and reveals the dominant
frequencies.
 Plot of angle(X[k]) vs. k is called the phase spectrum and provides information
about timing relationships.

4. Advantages and Limitations:

 Efficiently analyzes finite-length signals.

 Computes both magnitude and phase information.
  Requires power of 2 samples for optimal efficiency.
 Limited resolution for analyzing continuous-time signals.

UNIT – 3

1. FIR Filters:

 Finite impulse response: Output depends only on a finite number of past input
samples.
 Advantages: Stable, linear phase, easy to design.
 Disadvantages: May require higher filter order for comparable performance
compared to IIR filters.

1.1 Window Method:

 Multiplies the ideal filter frequency response with a window function to reduce
ringing in the stopband.
 Popular windows: Rectangular, Hamming, Hanning, Blackman.
 Trade-off: Window width vs. transition bandwidth.

1.2 Park-McClellan Remez Exchange Algorithm:

 Optimizes the filter coefficients to minimize the maximum deviation from the
desired frequency response within specified passband and stopband regions.
 More computationally expensive than window method but provides better
performance.

2. IIR Filters:

 Infinite impulse response: Output depends on both past and present input
samples and past outputs.
 Advantages: Lower filter order for comparable performance compared to FIR
filters, efficient for realization with analog circuits.
 Disadvantages: Potential for instability, non-linear phase at high frequencies.
2.1 Butterworth Approximation:

 Maximally flat magnitude response in the passband.

 Gentle roll-off in the stopband.
 Simple design equations and stable for all orders.

2.2 Chebyshev Approximation:

 Equal ripple magnitude response in the passband or stopband.

 Steeper roll-off than Butterworth filter at the expense of passband ripple or
stopband attenuation.

2.3 Elliptic Approximation:

 Ripple in both passband and stopband.

 Steeper roll-off than both Butterworth and Chebyshev filters for the same filter
order.
 More complex design and potential for instability for certain parameter
choices.

3. Filter Types:

 Low-pass: Attenuates frequencies above a certain cut-off frequency.

 High-pass: Attenuates frequencies below a certain cut-off frequency.
 Band-pass: Passes a specific range of frequencies between two cut-off
frequencies.
 Band-stop: Attenuates a specific range of frequencies between two cut-off
frequencies.

4. Design Considerations:

 Sampling rate: Affects the frequency range of the filter and the choice of filter
type.
 Passband and stopband specifications: Define the desired frequency
response of the filter.
 Filter order: Determines the complexity of the filter and its ripple
characteristics.
 UNIT – 4

1. Effect of Finite Register Length in FIR Filter Design:

 Real-world implementations involve limited word lengths (e.g., 8-bit, 16-bit) for
coefficients and computations.

 Consequences:

o Quantization noise: Rounding errors introduce noise into the filter

output.
o Limit cycles: Oscillations can occur due to overflow or underflow,
distorting the output.
o Coefficient sensitivity: Small changes in coefficients due to quantization
can significantly affect the frequency response.
 Mitigation strategies:

o Scaling: Normalize coefficients to avoid overflow/underflow.

o Word-length optimization: Choose appropriate word length for
acceptable noise levels.
o Dithering: Add random noise to reduce quantization error.

2. Parametric and Nonparametric Spectral Estimation:

 Spectral estimation: Estimating the power spectral density (PSD) of a signal to

understand its frequency content.

 Parametric methods:

o Assume a specific model for the signal (e.g., AR, ARMA).

o Estimate the model parameters and derive the PSD from the model.
o Advantages: Efficient for low-order models, good for stationary signals.
o Disadvantages: Sensitive to model mismatches, may not be suitable
for non-stationary signals.
 Nonparametric methods:
 o Do not assume any specific model for the signal.
o Estimate the PSD directly from the data using techniques like:
 Periodogram: Simple but noisy estimate.
 Welch's method: Averages multiple periodograms to reduce
noise.
 Maximum Entropy Method (MEM): Combines periodogram with
smoothness constraints.
o Advantages: More robust to model mismatches, suitable for non-
stationary signals.
o Disadvantages: May require more data compared to parametric
methods, may not capture sharp spectral peaks.

3. Introduction to Multirate Signal Processing:

 Deals with signals whose sampling rates are changed (up/down sampling).

 Applications:

o Efficient filtering of band-limited signals.

o Speech/audio compression and resampling.
o Image scaling and interpolation.
 Key concepts:

o Decimation: Reducing the sampling rate by an integer factor (L).

o Interpolation: Increasing the sampling rate by an integer factor (L).
o Anti-aliasing filters: Used to remove unwanted frequencies before
decimation to avoid aliasing.

4. Applications of DSP:

 Vast array of applications across various fields:

o Communication: Signal processing for wireless communication,
audio/video coding, error correction.
o Signal processing: Image/video processing, speech/audio processing,
radar/sonar systems.
o Control systems: Feedback control systems, robotics, motor control.
o Biomedical engineering: ECG/EEG analysis, medical imaging,
prosthetics.
Written by Engineer AD Sooryavanshi