EEE4510: Digital Signal Processing

Course Description

EEE 4510 Digital Signal Processing is the standard undergraduate introduction to processing signals in discrete time. It is where a student who has learned continuous-time signals and systems moves to sampled data: how a continuous signal becomes a sequence of numbers without losing what matters, what the frequency content of that sequence means, how to build filters that shape it, and how to compute all of this efficiently enough to run in real time.

The Statewide Course Numbering System titles this course Introduction to Digital Signal Processing and describes it as covering “sampling and data acquisition, design of simple digital filters, programming and hardware implementation, audio applications, basic spectrum analysis using FFT.” The FAMU–FSU College of Engineering states the content in more detail as “sinusoids, periodic signals, and Fourier spectra; sampling, aliasing, impulse response, convolution, FIR and IIR filters, Z-transforms, Discrete Fourier transform, and the FFT algorithm.”

This is the most widely offered course in the EEE prefix — five Florida institutions carry it: Florida A&M University, Florida Atlantic University, Florida International University, Florida Polytechnic University and Florida State University. That breadth means the content is well settled, and the guide below can be more definite than most. Four of the five title it plainly as digital signal processing or an introduction to it.

⚠ One institution has moved the course. FIU carries EEE 4510 under the title Introduction to Digital Signal Processing with AI. The underlying signal-processing content is the same, but the treatment integrates machine-learning methods applied to signals. A student choosing between institutions, or transferring, should expect FIU’s version to spend time on learned representations that the others spend on classical filter design. The AI Integration section returns to this.

Learning Outcomes

Required Outcomes

• Represent discrete-time signals and systems mathematically, and classify systems as linear, time-invariant, causal and stable.
• Apply the sampling theorem: determine the Nyquist rate for a given signal, predict aliasing when it is violated, and specify anti-aliasing and reconstruction filtering.
• Compute the response of an LTI system by convolution, and relate the impulse response to the system’s difference equation.
• Use the z-transform and its region of convergence to analyse discrete-time systems; find poles and zeros and relate their placement to stability and to frequency response.
• Compute and interpret the discrete-time Fourier transform and the discrete Fourier transform, and distinguish clearly between them.
• Explain and apply the FFT, including its computational advantage over the direct DFT, and interpret an FFT magnitude spectrum including the effects of windowing, leakage and zero-padding.
• Design FIR filters by windowing and by frequency sampling, and explain why FIR filters can be exactly linear phase.
• Design IIR filters by transformation from analogue prototypes (Butterworth, Chebyshev) using the bilinear transform or impulse invariance, and state the trade-off against FIR designs.
• Realise a filter in a chosen structure (direct form, cascade, parallel) and explain why the structure affects numerical behaviour.
• Implement and verify a signal-processing algorithm in software, and interpret the result against theoretical prediction.

Optional Outcomes

• Analyse finite word-length effects — coefficient quantisation, round-off noise, limit cycles and overflow — in fixed-point implementations.
• Apply multirate techniques: decimation, interpolation, polyphase decomposition and sample-rate conversion.
• Implement algorithms on a DSP processor or FPGA in real time, subject to a throughput or latency budget.
• Estimate power spectral density for random signals and distinguish deterministic from statistical spectral analysis.
• Apply adaptive filtering (LMS and related algorithms) to noise cancellation or system identification.
• Apply DSP methods to a specific application domain — audio, speech, communications, biomedical signals, radar or images.
• Apply machine-learning methods to signal classification and feature extraction (emphasised at FIU).

Major Topics

Required Topics

• Discrete-time signals and systems — sequences, the unit sample and unit step, basic operations, system properties, difference equations.
• Sampling and reconstruction — the sampling theorem, Nyquist rate, aliasing, anti-aliasing filters, zero-order hold and reconstruction, analogue-to-digital and digital-to-analogue conversion.
• Convolution and the impulse response — linear convolution, circular convolution, and the relationship between them.
• The z-transform — definition, region of convergence, properties, inverse transform by partial fractions, transfer functions, poles and zeros, stability and causality.
• Frequency-domain analysis — the DTFT, frequency response magnitude and phase, group delay, and the geometric interpretation of pole and zero placement.
• The discrete Fourier transform — definition and properties, the relationship to the DTFT, spectral leakage, windowing and window functions, zero-padding and resolution.
• The fast Fourier transform — decimation in time and in frequency, radix-2 algorithms, computational complexity, and practical use for spectrum analysis and fast convolution.
• FIR filter design — linear phase conditions and symmetry, window method, frequency-sampling method, and the Parks–McClellan equiripple approach.
• IIR filter design — analogue prototypes, impulse invariance, the bilinear transform and frequency warping, and frequency transformations between filter types.
• Filter structures — direct forms I and II, cascade and parallel realisations, lattice structures, and transposition.
• Implementation — programming filters and transforms in MATLAB or Python, and the transition from floating-point prototype to a practical implementation.

Optional Topics

• Finite word-length effects, fixed-point arithmetic, quantisation noise and limit cycles.
• Multirate signal processing and filter banks; introduction to wavelets.
• Real-time implementation on DSP hardware or FPGA, including memory and throughput constraints.
• Random signals, correlation, and power spectrum estimation (periodogram, Welch’s method).
• Adaptive filters — LMS, normalised LMS, RLS — and applications in echo and noise cancellation.
• Application modules in audio and speech, digital communications, biomedical signal processing, or image processing.
• Machine learning on signals: feature extraction, classification, and neural approaches to denoising and recognition (FIU).

Resources & Tools

• Discrete-Time Signal Processing (Oppenheim and Schafer) is the reference standard for the subject and the most commonly cited text, though its full depth exceeds a one-semester undergraduate course. Digital Signal Processing: Principles, Algorithms and Applications (Proakis and Manolakis) and DSP First (McClellan, Schafer and Yoder) are the two most commonly adopted undergraduate texts; Understanding Digital Signal Processing (Lyons) is widely recommended as a companion for intuition.
• MATLAB with the Signal Processing Toolbox is the dominant environment in Florida engineering programmes; filterDesigner, freqz, fft and fvtool are used routinely in assignments.
• Python with NumPy, SciPy (scipy.signal) and Matplotlib is increasingly used alongside or instead of MATLAB, and is the environment most likely to be encountered in industry outside defence.
• DSP hardware — Texas Instruments C6000 and C2000 family processors and their development kits, and Analog Devices SHARC processors, where a course includes a real-time laboratory component.
• FPGA tools — Vivado or Quartus with DSP-oriented IP, where implementation is taken to hardware.
• Reference material: the IEEE Signal Processing Society, the IEEE Transactions on Signal Processing, and the IEEE Signal Processing Magazine, which publishes accessible tutorial articles suitable for undergraduates.

Career Pathways

• DSP engineer and signal processing engineer (SOC 17-2071, Electrical Engineers) — the direct path, and one of the better-compensated specialisations in electrical engineering.
• Communications systems engineer — modulation, equalisation and synchronisation are all DSP problems in modern systems.
• Radar and sensor engineer (SOC 17-2071) — particularly relevant in Florida, where radar and electronic warfare work is concentrated.
• Audio and acoustics engineer; biomedical signal processing engineer (SOC 17-2031, Bioengineers and Biomedical Engineers).
• Machine learning engineer working on sensor, speech or image data (SOC 15-2051, Data Scientists) — the route FIU’s version of the course points at most directly.
• Embedded software engineer in signal-processing-heavy products.
• Florida employers include L3Harris (Melbourne and Palm Bay — communications, electronic warfare and radar, and one of the largest employers of Florida DSP graduates), Lockheed Martin (Orlando), Northrop Grumman (Melbourne and St. Augustine), Raytheon/RTX (Largo), Motorola Solutions (Plantation), Siemens, Medtronic and Johnson & Johnson MedTech (Jacksonville and Palm Beach Gardens) for biomedical instrumentation, and the naval research and surface warfare activity around Panama City.

Special Information

Prerequisites

The universal prerequisite is a continuous-time signals and systems course — EEL 3135 at the FAMU–FSU College of Engineering and at most other Florida institutions. Students are expected to arrive comfortable with the Fourier and Laplace transforms, convolution, complex arithmetic and linear differential equations. The SCNS record lists EEL 4656 as a prerequisite, which does not reflect current practice at the institutions offering the course; use your own institution’s catalog.

⚠ The mathematics that actually causes trouble is not always named in the prerequisite. This course leans hard on complex numbers — magnitude and phase, Euler’s identity, and the geometry of poles and zeros in the complex plane — and on comfort with summations and series. Students who can pass a differential equations course but are shaky on complex arithmetic tend to meet the difficulty in week three, at the z-transform, and conclude the course is harder than it is. Reviewing complex algebra and the geometric meaning of \(e^{j\omega}\) before the term starts is worth more than any other preparation.

Course-code variation across Florida

Digital signal processing is taught under several numbers in the state, and EEE 4510 is only one of them. The University of Florida carries EEE 4511C Real Time Digital Signal Processing Applications (4 credits, laboratory-based) rather than a course at this number, and graduate-level treatments appear as EEE 5502 and EEE 6502. Some programmes deliver the material under an EEL number instead. Credit transfers, but a receiving programme matching prerequisites by number may not recognise a different number without a substitution request — carry the syllabus.

⚠ Institutional variation in emphasis

All five institutions cover the same core. The differences are of emphasis and are worth knowing before you register:

• FIU — Introduction to Digital Signal Processing with AI. Classical material plus machine-learning methods applied to signals.
• FAMU and FSU (joint College of Engineering) — a classical treatment running from Fourier spectra through Z-transforms to FIR/IIR design and the FFT.
• FAU — titled as an introduction, consistent with the statewide description.
• Florida Polytechnic — a strongly applied, implementation-oriented programme in general; expect more hands-on work.

A practical test for which version you are in: if the assessed work is filter design against a specification and FFT-based spectral analysis, it is the classical course; if it includes training a classifier on signal features, it is the AI-integrated version.

Position in the curriculum

EEE 4510 is a senior-level elective in most electrical and computer engineering programmes and a required course in some. It is the foundation for graduate DSP (EEE 5502, EEE 6502), for adaptive signal processing, for radar (EEE 4550, which lists a signals prerequisite alongside electromagnetics), for communications systems, and for image and speech processing. Where a programme offers a signal-processing concentration, this is its entry point.

Difficulty and time commitment

This is one of the more demanding electives in the electrical engineering curriculum — comparable to electromagnetics in the effort it asks for, and for a similar reason: the mathematics is not individually hard, but it is abstract, cumulative, and unforgiving of gaps. Plan on eight to twelve hours a week, and expect the programming assignments to take longer than they look. Students who work the derivations by hand before reaching for MATLAB do substantially better; the tool will produce a filter whether or not you understand it.

Articulation and transfer

SCNS records EEE 4510 as guaranteed to transfer to an institution offering the same course, and with five institutions carrying the number that guarantee has real reach — unusually good coverage for an upper-division elective. The course is upper-division, carries no general-education or Gordon Rule designation, and all five institutions carry it at 3 credits.

FE exam relevance

The NCEES Fundamentals of Engineering (Electrical and Computer) exam includes Signal Processing as a named topic area, covering sampling, discrete-time signals and systems, convolution, difference equations, Z-transforms, filters and the discrete Fourier transform. The overlap with this course is close to exact, and it is one of the more efficient FE topics to prepare from coursework rather than from a review manual.

AI Integration

Signal processing is one of the fields where machine learning has genuinely displaced part of the classical toolkit, which makes the question here substantive rather than decorative. FIU has gone as far as retitling the course Introduction to Digital Signal Processing with AI, and that is a reasonable read of where the discipline is going.

Where AI methods have genuinely taken over. Speech recognition, speech enhancement, audio source separation, image denoising and super-resolution, and modulation classification are all tasks where learned models now outperform hand-designed signal-processing pipelines, often substantially. A student who leaves this course believing that a carefully designed Wiener filter is the state of the art in speech enhancement has been misinformed.

⚠ Where the classical material remains non-negotiable. Sampling and aliasing are physics, not modelling choices — no amount of learning recovers information destroyed by sampling below the Nyquist rate. Filters in the signal path of a safety-critical or regulated system must have analysable stability and phase behaviour, which a neural network does not provide. Real-time and low-power systems still run FIR and IIR filters because the computational budget allows nothing else. And a learned model trained on one sensor and deployed on another fails in ways that a pole-zero plot would have predicted.

⚠ The failure mode that coincides with this course’s own subject. The characteristic error of an AI tool asked a DSP question is to produce code that runs and returns a plausible spectrum that is wrong — an FFT without a window applied, so leakage is misread as signal content; a filter designed to a normalised frequency the model has scaled incorrectly against the sampling rate; a magnitude spectrum plotted over the full range so the mirrored negative frequencies are interpreted as real peaks. These are exactly the misreadings the course exists to prevent, and none of them raises an error. Verifying against a known test signal — a pure sinusoid at a known frequency, an impulse — catches all three in about a minute, and that habit is one of the more durable things to take from the course.

The engineer’s responsibility. A spectrum is a claim about a physical signal, and the engineer signs for it. If you cannot say what the frequency axis units are, what window was applied, and what the resolution is, you are not in a position to defend the plot — whoever or whatever produced it.

Academic integrity. Institutional policies on AI-assisted work differ, and instructors of the same course differ. Derivations are typically expected to be your own; code assistance may be permitted with disclosure. Ask before you rely on a tool.