DSP Applications: Audio, Radar, Biometrics, and BCI

Why Signals Are Processed

Signals carry both useful information and unwanted noise. Signal processing exists to extract, enhance, store, and transmit the useful part. There are two ways to do it:

Approach	How	Examples
Analog signal processing	Continuous circuits: resistors, capacitors, op-amps	Passive filters, analog radio
Digital signal processing	Sample, quantize, then compute	Everything modern

DSP is preferred because a program can be changed without touching hardware, and because digital arithmetic has no component tolerance drift.

---

The Five Core Application Areas (CSE 409)

1. Audio Processing and Speech Recognition

Speech recognition is a two-stage DSP pipeline:

Microphone → Pre-emphasis → Framing/Window → Feature Extraction → Feature Matching → Text
             (boost highs)   (20-30 ms)       (MFCC, LPC, mel)     (HMM/DNN)

• Feature extraction turns each short frame into a compact vector (MFCCs are the classic choice).
• Feature matching compares that vector sequence against acoustic models.

Other audio jobs: noise suppression, echo cancellation, equalization, compression (MP3/AAC), pitch shifting.

2. Echo Location (Radar and Sonar)

A wave is bounced off a remote object; the echo reveals distance (delay), speed (Doppler shift), and size/shape (amplitude and signature).

• Radar: electromagnetic pulses.
• Sonar: acoustic pulses underwater.
• Ultrasound: the same idea at MHz frequencies inside the body.

DSP performs matched filtering (pulse compression), clutter rejection, beamforming, and data compression of the returns.

3. Pattern Recognition

“The act of taking in raw data and taking an action based on the category of the data.”

Pattern recognition classifies data using a priori knowledge or statistical information extracted from patterns. DSP supplies the feature vector; the classifier supplies the decision.

Examples: handwritten digit recognition, ECG arrhythmia detection, fault diagnosis from vibration, image classification.

4. Biometrics

Biometrics identifies a person from intrinsic physical or behavioural traits: face, voice, fingerprint, iris, handwriting, gait.

Each trait is a signal:

Trait	Signal	DSP stage
Face	2-D image	Enhancement, segmentation, feature extraction
Voice	1-D waveform	MFCC, speaker modelling
Fingerprint	2-D ridge image	Gabor filtering, minutiae extraction
Gait	Accelerometer time series	Periodicity and spectral features

Exam answer: Yes, biometrics is a DSP application. Every biometric system measures a physical quantity, conditions it, extracts features, and matches it.

5. Brain-Computer Interface (BCI)

A BCI provides communication between a brain and a computer from measurements of brain activity (EEG, ECoG, implanted arrays) with no muscle motion involved.

DSP chain: artifact removal → band-power extraction (alpha/beta/gamma) → spatial filtering (CSP) → classifier → control command.

---

Beyond the Lecture List

Domain	DSP task
Medical imaging	CT/MRI reconstruction, denoising, edge enhancement
Communications	Modulation, equalization, channel coding, OFDM
Consumer electronics	Noise-cancelling headphones, image stabilization, voice assistants
Seismology	Deconvolution, velocity analysis, event detection
Finance	Trend/seasonality extraction, volatility filtering
Industrial control	Sensor conditioning, vibration monitoring

---

Applications Mapped to Course Topics

Application	Topics used
Speech recognition	Sampling, framing, windowing, FFT, cepstrum, LPC
Radar	Correlation, matched filtering, FFT, Doppler spectral analysis
Noise cancellation	Adaptive filters (LMS/RLS), Wiener filtering
Audio compression	Filter banks, QMF, psychoacoustic models
Image processing	2-D DFT/DCT, 2-D filtering, wavelets
BCI	Bandpass filtering, spatial filtering, spectral estimation

---

Key Takeaways

1. DSP is not one algorithm; it is a pipeline of sampling → conditioning → transformation → decision.
2. The same four tools — convolution, transforms, filters, statistics — appear in every application.
3. Applications are the reason the theory exists: pick the tool that matches the signal’s structure.