Application of Intelligent Tutoring Systems in Digital Signal Processing Teaching: Empowering Educational Practice with Large Language Models

Digital Signal Processing (DSP) is a core course across multiple disciplines. Traditional teaching faces challenges such as students' difficulty in understanding, insufficient personalized tutoring, and disconnect from practice. This project explores the application of Large Language Models (LLMs) in DSP teaching and builds an intelligent tutoring system to enhance personalized learning experiences and teaching effectiveness. The system combines LLMs' natural language understanding and knowledge integration capabilities with structured knowledge in the DSP domain to provide 24/7 personalized support, covering theoretical tutoring, programming practice, and other links. It aims to address the pain points of traditional teaching and provide practical references for AI-empowered education.

DSP involves abstract concepts such as Fourier transform and filter design, which are difficult for beginners to understand intuitively. Traditional teaching relies on classroom lectures and textbooks, which cannot meet personalized needs: students with weak foundations need supplementary prerequisite knowledge, advanced students need more practical guidance, and teachers have limited energy to provide one-on-one tutoring. In hands-on sessions, error detection and code debugging in programming (MATLAB/Python) and hardware platform operations often become learning barriers. Intelligent Tutoring Systems (ITS) are designed to address these pain points and provide personalized learning support via AI.

The system adopts a modular architecture: the knowledge base module integrates core DSP knowledge (textbooks, papers, examples, etc.); the dialogue engine handles user interactions and manages context; personalized modeling tracks learning trajectories to build user profiles; the code assistance module supports programming practice; the evaluation and feedback module embeds diagnosis into the learning process. For DSP knowledge modeling, a hierarchical concept map is built (e.g., Signals and Systems → Transform Domain Analysis → DFT/FFT), and a library of common questions (e.g., confusing linear and circular convolution), a library of visual descriptions (spectrum diagrams, zero-pole diagrams), and a library of code patterns (typical algorithm implementations) are collected to help LLMs understand the domain knowledge structure.

The system implements multiple personalized strategies: scaffolding tutoring (hierarchical support with gradually reduced guidance), Socratic questioning (guiding active thinking), error diagnosis and correction (feedback targeting specific errors), and learning path recommendation (adapting content based on knowledge mastery). Practical support includes: code explanation (line-by-line description of functions and principles), debugging assistance (analyzing error causes), result verification (checking the rationality of experimental results), and experimental design guidance (helping plan experimental schemes).

The project evaluates the system's effect through multiple dimensions: 1. Learning performance comparison: The experimental group uses the system for assistance, while the control group uses traditional methods; compare exam scores, homework quality, etc. 2. Learning process analysis: Record behavioral data such as usage frequency, question topics, and improvement status. 3. Satisfaction survey: Collect students' feedback on the system's ease of use and the usefulness of answers. 4. Teacher feedback: Observe changes in classroom performance, evaluate whether the system reduces the burden of answering questions and identifies common problems.

The system has limitations: accuracy issues (LLMs may generate incorrect explanations), insufficient deep understanding (superficial analogies), and limited personalization boundaries (difficulty in perceiving implicit cognitive states). Future directions include: enhancing multimodal interaction (voice, handwriting), supporting collaborative learning, integrating with virtual experimental environments, and improving knowledge tracking algorithms.

This project provides implications for AI education: 1. Importance of domain adaptation: General-purpose LLMs need to be combined with domain knowledge bases and teaching strategies. 2. Human-machine collaboration: AI takes on repetitive tutoring, while teachers focus on creative teaching. 3. Continuous iteration: Optimize the system using teaching data. 4. Ethical considerations: Protect student data, avoid over-reliance, and ensure educational equity. In summary, AI tutoring systems are expected to become an auxiliary force in education, helping students overcome obstacles in learning complex knowledge.