Mastering Digital Signal Processing: Key Concepts from an ECE113 Midterm Exam

Introduction: Why DSP Matters in 2026

In 2026, digital signal processing (DSP) powers everything from AI voice assistants to 5G communications and real-time gaming audio. Understanding core DSP concepts is crucial for electrical and computer engineering students. This tutorial breaks down key topics from an ECE113 midterm exam, helping you grasp discrete-time Fourier transforms, filter realizations, spectral analysis, and sample rate conversion. Whether you're studying for an exam or brushing up on fundamentals, these explanations connect theory to real-world applications.

1. Discrete-Time Sinusoids and Periodicity

Statement: Discrete-time sinusoids are always periodic in time. False. A discrete-time sinusoid cos(ω₀n) is periodic only if ω₀/2π is a rational number. For example, cos(0.5n) is not periodic because 0.5/(2π) is irrational. This is a common pitfall in DSP exams.

2. DTFT Periodicity in Frequency

Statement: Discrete-time Fourier Transforms are always periodic in frequency. True. The DTFT X(e^{jω}) is periodic with period 2π because e^{j(ω+2π)n}=e^{jωn}. This property is fundamental for digital filter design and spectral analysis.

3. Direct Form I vs. Direct Form II Overflow

Statement: A finite-precision Direct Form I realization has a higher chance for internal overflow compared to Direct Form II. True. In Direct Form I, the zeros (feedforward) are implemented before the poles (feedback), causing larger intermediate signal values. Direct Form II reverses this order, reducing internal dynamic range. This is critical in fixed-point implementations for embedded systems like hearing aids.

4. Zero-Padding and Spectral Leakage

Statement: Zero-padding always leads to smaller spectral leakage. False. Zero-padding interpolates the DTFT but does not reduce leakage; it only provides a finer frequency grid. Leakage depends on the window function. For instance, a rectangular window causes significant leakage regardless of zero-padding.

5. Phase of DTFT for Even-Symmetric Real Signals

Statement: The phase of the DTFT of an even-symmetric real-valued signal is always 0 or 180°. True. For real and even signals, the DTFT is real and even, so the phase is 0 (when positive) or π (when negative). This is used in linear-phase filter design.

6. Downsampling and Aliasing Conditions

Statement: Downsampling a lowpass sequence by factor D will never cause aliasing if the sequence has no energy in [π/D, π]. True. Downsampling expands the spectrum by factor D, so the original band must be limited to π/D to avoid overlap. This is why anti-aliasing filters are used before decimation in audio sample rate converters.

7. Upsampling and Aliasing

Statement: Upsampling by factor I may lead to aliasing if the sequence has energy in [π/I, π]. False. Upsampling inserts zeros and compresses the spectrum, creating images at multiples of 2π/I. Aliasing occurs only if the original signal has energy beyond π/I, but the images are removed by an interpolation filter. The statement incorrectly uses "aliasing" for imaging.

8. DFT Representation of Finite-Length Sequences

Statement: Any discrete-time sequence of length N or less can always be represented by its N-point DFT. True. The DFT is a one-to-one mapping for sequences of length ≤ N. This is the basis for OFDM in Wi-Fi and 5G.

9. Sampling Rate for Bandpass Signals

Statement: The minimum sampling rate to avoid aliasing for a real-valued bandpass signal with bandwidth B is always 2B. False. Bandpass sampling can use a rate as low as 2B, but the sampling frequency must also satisfy constraints to avoid spectral overlap. The minimum rate is 2B only if the band is positioned correctly, e.g., from B to 2B.

10. Circular vs. Linear Convolution

Statement: N-point circular convolution equals linear convolution in [0,N-1] if N ≥ max(N1,N2). False. The condition is N ≥ N1 + N2 - 1 to avoid time-domain aliasing. For example, convolving length-3 and length-4 sequences requires N ≥ 6 for equivalence.

11. LCCDE and Direct Form Structures

Consider the second-order LTI system: y(n) - 0.1y(n-1) - 0.72y(n-2) = 5x(n-1). The characteristic equation is r² - 0.1r - 0.72 = 0, with roots r = 0.9 and r = -0.8. The homogeneous response is y_h(n) = C₁(0.9)^n + C₂(-0.8)^n. Since |0.9|<1 and |-0.8|<1, the system is BIBO stable. For input x(n) = (1/5)^n u(n), the complete response includes both natural and forced modes. The Transposed Direct Form II structure minimizes memory and is common in FPGA implementations for real-time audio effects.

12. DTFT and Frequency Response

Given an impulse response h(n), the frequency response H(ω) is the DTFT of h(n). For a causal system, h(n)=0 for n<0. Stability requires absolute summability. Using DTFT properties, you can find H(ω) in closed form. For example, if h(n) = a^n u(n), then H(ω)=1/(1-ae^{-jω}). The steady-state response to a sinusoidal input is H(ω₀) times the input at that frequency.

13. Autocorrelation and DTFT Properties

An autocorrelation sequence must be even, maximum at zero, and positive semidefinite. The pulse sequence u(n)-u(n-N) is not valid because it is not even. The triangle sequence Λ(n/(2N+1)) is valid as it is the autocorrelation of a rectangular pulse. For a delayed version y(n)=x(n-n₀), r_{yy}(n)=r_{xx}(n). For modulation y(n)=e^{jω₀n}x(n), r_{yy}(n)=e^{jω₀n}r_{xx}(n). These properties are used in spectrum estimation.

14. DFT Spectral Analysis in MATLAB

To improve spectral representation of a two-tone signal with AWGN, you can: (1) increase the FFT length (zero-padding) for finer frequency resolution; (2) use a window function like Hamming or Blackman to reduce spectral leakage; (3) average multiple FFTs (Welch's method) to reduce noise variance. For example, with Fs=8.5 kHz and 10 dB SNR, using a Blackman window and 1024-point FFT with 50% overlap can reveal the two tones clearly. This technique is similar to how Spotify's equalizer visualizes music frequencies.

15. Sample Rate Conversion

When two signals sampled at different rates need to be combined, you must upsample, filter, and downsample to a common rate. For example, converting a 44.1 kHz audio signal to 48 kHz involves interpolation and decimation. Anti-aliasing filters are critical to avoid aliasing. This is used in modern audio interfaces and video processing.

Conclusion

Mastering these DSP concepts not only helps in exams like ECE113 but also prepares you for careers in AI, communications, and consumer electronics. Practice with MATLAB and explore real-world signals to solidify your understanding.