CENG 420 Design of Digital Signal Processing Systems

1
CENG 420 Design of Digital Signal Processing
Systems

• Dr. Brian T. Hemmelman

2
Introduction

• Historically, the signals and information most
  engineering dealt with were always analog
  (voltage, speed, etc.)
• With the advent of the digital computer, new
  possibilities opened up.
• Mimic analog signal conditioning using digital
  data
• DSP can perform some tasks better than analog
  methods, and can actually do tasks analog signal
  processing cant (image processing, data
  compression, etc.)

3
Typical DSP Signals

• Speech and voice
• Sound, music, audio
• Image, video
• Sonar, radar
• Satellite sensors (Infrared, UV, CO2, )
• Biomedical (ECG, EEG, EMG, MRI, ultrasound, )

4
DSP Applications

• Image Processing Robotic vision, FAX, satellite
  weather
• Instrumentation Spectrum analysis, noise
  reduction
• Speech Audio Speech recognition, equalization
• Military Radar processing, missile guidance
• Telecommunications Echo cancellation, video
  conferencing, VoIP
• Biomedical ECG analysis, patient monitoring
• Consumer Electronics Cell phones, set top box,
  video cameras

5
DSP Applications
6
Advantages of DSP

• Guaranteed Accuracy Accuracy only limited by
  bit length
• Perfect Reproducibility No component
  tolerances, no component drift due to temperature
  or age
• Greater Flexibility Functions and algorithms
  can be changed through software
• Superior Performance Adaptive filtering, linear
  phase response
• Some Data Naturally Digital Images, computer
  files

7
Disadvantages of DSP

• Speed and Cost ADC/DAC, uProc
• Design Time Can be tricky
• Finite Word Length Issues

8
Key DSP Operations

• Convolution
• Correlation
• Filtering
• Transformations
• Modulation

9
Convolution

• Many uses, but a common use is determining a
  systems output if system input and system
  impulse response is known. For continuous system

10
Discrete Convolution

• We may however have a computer sampling a signal
  so that we have discrete data.
• So instead of continuous integration process we
  have discrete summation.

• Practically speaking though we would have finite
  sequences x(n) and h(n) of lengths N1 and N2
  respectively, so this is then

11
Discrete Convolution

• Note that this is a series of multiplies-followed-
  by-additions, so that this operation is
  fundamentally a Multiply-Accumulate or MAC.
• DSP systems are often benchmarked by the number
  of MACs per second they perform.
• DSP chips (and many FPGAs) have special internal
  architectures that help perform MACs more
  efficiently.

12
Correlation

• Correlation is essentially the same as
  convolution (from a computational standpoint).
  You just dont flip anything.
• Instead of describing system output, correlation
  tells us information about the signals.

13
Correlation

• Cross-correlation function
• Tells you a measure of similarities between two
  signals.
• Application Identifying radar return signals

14
Correlation
15
Correlation

• More than one definition of cross-correlation
• One definition for two N-length sequences

16
Correlation

• Autocorrelation function
• Correlate a signal with itself.
• Helps find periodicity in signals.

17
Correlation
18
Digital Filters

• High-pass, low-pass, bandpass, etc.
• Basically same idea as analog filters.
• FIR filter form

• x(n) is input
• y(n) is output
• h(k) are filter coefficients

19
Digital Filters
20
Discrete Transformations

• Often we want to find the frequency components of
  a signal and/or need to go back and forth between
  the time and frequency domain.
• The most common technique is the DFT (not DTFT)

21
Compact Disc Example

• Both channels are sampled and mixed for storage.
• Dual channel data is encoded with a Reed-Solomon
  encoder to fix burst errors (e.g. scratches).
• To be more suitable for optical storage
  Eight-to-Fourteen Modulation (EFM) is used.

22
Compact Disc Example

• Optical signal picked off CD and demodulated.
• Reed-Solomon error correction and concealment is
  performed.
• Dual-channel data is oversampled (4x) and split
  into left and right channels.
• Digital data is converted to analog voltage and
  lowpass filtered.

23
Fetal ECG Monitoring

• Track babys heart activity through electrical
  potentials on body surface.
• Some key features
• ST segment
• Ratio of T amplitude to QRS amplitude
• Width of QRS complex

24
Fetal ECG Monitoring
Cardiotocogram (CTG)
25
Fetal ECG Monitoring

• These signals are quite susceptible to noise from
  power supplies, muscle artifacts, poor
  connections, etc.
• Typical approach
• Remove 50/60 Hz noise
• Locate R-wave (most noticeable feature)
• Remove baseline shift
• Filter muscle noise
• Identify ST and PR segments

26
Fetal ECG Monitoring