DTFT Overview

The Discrete Time Fourier Transform (DTFT) is the foundation of frequency analysis for discrete-time signals. It maps an infinite-length discrete sequence into a continuous, periodic frequency spectrum, making it essential for understanding how digital systems process signals across all frequencies.

In GATE and university exams, DTFT appears in problems involving frequency response of LTI systems, filter design, and the derivation of the DFT itself. Understanding DTFT deeply makes all other transforms easier to grasp.

Core Concept Explanation

A discrete-time signal x[n] exists only at integer time indices. When you want to analyze what frequencies this signal contains, you compute the DTFT. Unlike the continuous Fourier Transform applied to analog signals, the DTFT accepts samples as input but produces a continuous output spectrum. This continuous spectrum is then periodic with period 2pi in the digital frequency variable omega (w).

The periodicity arises because discrete-time systems cannot distinguish a sinusoid at frequency w from one at w + 2pi. Both produce identical sample sequences. This is a fundamental consequence of sampling, not a limitation of the transform itself. Every property of the DTFT flows directly from this underlying periodicity.

The DTFT exists for a signal x[n] when the sequence is absolutely summable, meaning the sum of absolute values of all samples is finite. Stable LTI systems with absolutely summable impulse responses always have a DTFT. For energy signals, the DTFT also exists in the mean-square sense.

The frequency response of an LTI system is simply the DTFT of its impulse response h[n]. This is why the DTFT is the gateway concept to filter design. A low-pass filter has a frequency response close to 1 for small w and close to 0 for w near pi, and this entire shape is described by the DTFT of its coefficients.

Mathematical Expression

The forward DTFT is defined as the sum of all samples multiplied by complex exponentials. For a sequence x[n], the DTFT is written as X(e^jw) = sum_{n=-inf}^{inf} x[n] e^{-jwn}. Here w is a continuous real variable representing digital frequency in radians per sample, ranging from -pi to pi for one period.

The inverse DTFT recovers the original sequence from the spectrum: x[n] = (1/2pi) integral_{-pi}^{pi} X(e^jw) e^{jwn} dw. This synthesis integral reconstructs each sample by integrating the spectrum weighted by the complex exponential at that time index.

For real-valued sequences, X(e^jw) exhibits conjugate symmetry: X(e^{-jw}) = X*(e^jw). This means the magnitude spectrum is even and the phase spectrum is odd. GATE problems often use this symmetry to simplify calculations involving only the positive frequency half.

Practical Understanding

In a digital filter implementation, the filter coefficients h[n] completely determine the frequency shaping. Computing H(e^jw) tells you exactly how each frequency component in the input is amplified or attenuated. A high-pass filter will show H(e^jw) near zero for small w and near one near w = pi.

The DTFT is not directly computable in practice because it requires infinite summation and produces a continuous result. The Discrete Fourier Transform (DFT) is the practical tool that samples the DTFT at N equally spaced frequencies, making it computable by machines. Every DFT bin is literally a sample of the DTFT at w = 2pi*k/N.

Given:
Sequence x[n] = {1, 2, 1} for n = 0, 1, 2 (zero otherwise)
Find X(e^jw) at w = 0 and w = pi

Why this formula applies:
DTFT definition: X(e^jw) = sum x[n] e^(-jwn)
Finite-length sequence so sum has finite terms.

Formula:
X(e^jw) = x[0]*e^0 + x[1]*e^(-jw) + x[2]*e^(-j2w)

Substitution at w=0:
X(e^j0) = 1*(1) + 2*(1) + 1*(1)

Calculation at w=0:
X(e^j0) = 1 + 2 + 1 = 4

Substitution at w=pi:
X(e^{jpi}) = 1*(1) + 2*(-1) + 1*(1)

Calculation at w=pi:
X(e^{jpi}) = 1 - 2 + 1 = 0

Final Answer:
X(e^j0) = 4 (DC gain = sum of all samples)
X(e^{jpi}) = 0 (highest frequency cancelled)
This sequence acts like a low-pass filter kernel.

Exam Tip: In GATE, X(e^j0) always equals the sum of all x[n] values. If asked to find DTFT at w=0, simply add all samples. At w=pi, alternate signs (+,-,+,-...) before adding.

Properties of DTFT

The time-shifting property states that delaying x[n] by n0 samples multiplies the DTFT by e^{-jwn0}. The magnitude spectrum stays unchanged while the phase spectrum gains a linear phase term. This is why FIR filters with symmetric coefficients have exactly linear phase.

The convolution property is perhaps the most powerful: convolution in time becomes multiplication in frequency. If y[n] = x[n] * h[n], then Y(e^jw) = X(e^jw) H(e^jw). This is the mathematical reason why filters work: they multiply out unwanted frequency components in the spectrum of the input.

The Parseval theorem for DTFT connects time-domain energy to frequency-domain energy: sum |x[n]|^2 = (1/2pi) integral |X(e^jw)|^2 dw. This ensures energy is conserved when you switch between domains, which is critical for power spectral analysis.

Quick Revision

• DTFT formula: X(e^jw) = sum x[n] e^(-jwn) from n = -inf to inf. Input is discrete, output is continuous and periodic.
• X(e^jw) is always periodic with period 2pi. One complete period from -pi to pi gives all unique frequency information.
• At w=0, X(e^j0) equals the sum of all x[n] samples. At w=pi, alternate signs before summing.
• Convolution in time = multiplication in frequency. This is the fundamental reason filters can be designed in the frequency domain.
• For real x[n], magnitude spectrum is even symmetric and phase spectrum is odd symmetric about w=0.
• DFT is just N equally spaced samples of the DTFT: X[k] = X(e^jw) evaluated at w = 2pi*k/N.
• Common exam trap: DTFT exists only if x[n] is absolutely summable. Non-stable systems may not have a DTFT but have a Z-transform.