Biomedical Signal Processing

Trịnh Quang Đức

 Curriculum
• Ch 1: Signal and Biomedical signal
• Ch 2: Discretizing
• Ch 3: Signal Analysis
• Ch 4: Signal Filtering
• Ch 5: Signal modeling
 Chapter 1
EOG
1.1 Bio-electrical Signal
EMG

EEG

ECG
1.2 Bio-optical Signal
SPO2
 1.3 Bio-acoustic signal

Heart sound

Lung sound

Abdomen sound
1.4 Bio-photon signal

Bức xạ sinh học

 1.5 Basics of biomedical signal
• Definition: The electromagnetic wave including biological
information and clinical conditions.

• Derivation of the active biomedical signal:

- Bio-electrical signal -> from metabolism in living cells.
- Bio-acoustic signal -> from movement of living organs.
- Bio-photon signal -> from oxidized bio-chemical reaction.

• Derivation of the inactive biomedical signal:

- Response from body under excitation of a physic energy (Electrics,
Photon, Magnet, acoustic, heat)
- Response from electromagnetic wave: X-ray, CT, MRI, PET, SPECT
- Response from acoustic wave: Ultrasound
 1.6 Approaches
• Profits of biomedical signal:
- Biomedical signal includes the significant
biological information to observe the clinical
conditions of living body-> clinical diagnosis.
- Response to excitation provide information of
living organs, tissues, then observe the clinical
conditions -> Clinical diagnosis and therapy.
• The roles of inactive biomedical signal
- Observe the clinical condition
Biomedical Imaging
- Observe the biological info.
1.7 Problems of biomedical diagnosis

Understand the Bio-signal extraction

interpretation of signal

Signal
processing

Characteristics of signal
1.8 Why biomedical signal processing?
• The sophistication of biomedical signal ->
multi wave forms, multi frequencies.
• The dynamic state of biomedical signal ->
stochastic, multi parameters.
• Measured signal contaminated from
environment noises.
 1.9 Biomedical Signal Processing
Methods
• Separating signal to individual specified signal
• Convert the signal from a domain to other
domain
• Analyzing signal from distribution of specified
signal
• Filter design to reduce the noises
• Check the efficiency of the filters through
analysis of signal before and after filtered
 Chapter 2:
Discretizing

ADC

Analog signal Digital signal

 2.1 Definition of signal format
• Analogue: Electromagnetic wave varying in time
continuously.
- Properties:
+ Time resolution -> infinity
+ Value resolution -> infinity
• Digital: Electromagnetic wave varying in time
with each period time.
- Properties:
+ Time resolution -> limited
+ Value resolution -> limited
2.2: Pros and Cons of Analog Signal
Pros
- Continuity -> continuous information (infinity)
- Allow to observe the variation in small time ->
The time length resolution is infinity.
• Cons
- Difficulties to noise removal
- Difficulties in complicated math functions
implement
- Difficulties in design signal analysis based on
electronic circuit.
2.3: Pros and Cons of Digital Signal
• Pros:
- Easy to process the signal based on math function
- Small effected from noise (or no noise) from other
sources after discretized.
- Easy to analysis, separate the signal based on math
function.
• Cons:
- The discrete values (discontinuity) in both value and
time. The limitation of the resolution
- Limitation of the accuracy.
- Real-time observation is limited
 2.3 Conversion tools

ADC (Analog to digital converter)

Analog Digital
Conversion
 2.4 parallel ADC

-Bit data of ADC provided parallelly

-The value of ADC is integer, corresponding to physic value
-Convert the value of ADC to physic value depending on ref+ và ref- voltage
to define the measurement range
 2.5 Serial ADC

-Data bit is arranged as serial type in a frame.

-Number of bit from the data frame usually high (24 bit to 32 bit)
-The data bit stored in shifted register to transfer from parallel to
serial.
 2.6 Sampling time
When the wheel turning fast, the direction
movement of the wheel reversing?

The errors of the information depends on sampling time

2.7 Sampling frequency
 2.8 Sampling theories
• A signal can rebuilt if and only if it is sampled with the sampling frequency
at least of double its own frequency. Nyquist theorem
2.9 How to sampling
 Ch. 3: Signal Analysis

Time Domain Frequency Domain

3.1: Advantage of Signal Conversion in
Frequency Domain
• Separation a complicated wave to be multi
component waves.
• Each component wave and number of
component wave contained in a signal wave
make its characteristics.
• Component basic wave is considered as sine
wave (Fourier) and included information
• Distribution of component waves from a signal
called spectrum. Spectrum of biomedical signal
contains biological information.
 3.2 Difficulties of numerical analysis
• Analog signal analysis based on electronic
circuit to perform complicated math
functions.
• Fourier transform.
• Fourier transform needed to be performed in
discrete domain.
• The transform from analog to discrete domain
called z transfrom
3.2.1 Signal Components

Solution

If the waveform includes many frequencies???

Complex wave
3.2.2 Domain Observation
 3.2.3 How to change?

Conversion
(Transform)

F(t) -> F(f)

 3.2.3 Terms of signal
• Time-domain signal -> waveform
• Frequency-domain signal -> spectrum
• Signal: the range of frequency should be
observed
• Noise: the range of frequency should be
eliminated
• Signal to noise ratio (SNR): Quality of
measurement SNR=20*log(Signal Level/Noise
Level)
 3.3 Z Transform
Laplace transform of f(t):
∞
F (s ) = f (t )e − st dt
∫0
Z transform of a side of x(n):
∞
X (z) = ∑ x(n )z − n
n =0
Z transform of both side of x(n):
∞
X (z) = ∑ x(n )z − n
n = −∞
3.4 Discrete Fourier Transform

Fourier transform of analog signal Fourier transform of digital signal

∞
X (re iω ) = ∑ x (n )( re iω ) −n , or
n = −∞
∞
X (re iω ) = ∑ x (n )r −n e −iωn , and if r = 1,
n = −∞
∞
X (e iω ) = X (ω ) = ∑ x (n )e −iωn
n = −∞
 3.5 Digital Fourier Transform & Fast
Fourier Transform FFT

FFT

Fourier Transform Fast Fourier Transform

DFT

Digital Fourier Transform Fast Fourier Transform

 3.6 Fourier Series
Signal contains many components. Fourier transform implements to a wave to
extract that wave to be components called Fourier series.
3.7 Example with square wave
 3.8 Square wave synthesis

Tổng hợp xung vuông

từ sóng sin

Nếu số lượng sóng sin hữu hạn -> hiện tượng Gibbs

Hiện tượng Gibbs

3.9 Phase analysis
3.10 Z Operation
3.11 Spectral Power Density
3.12: DFT basic function
3.13: Digital Fourier Transform
3.14: Digital Fourier Integral
3.15: Basic function of Digital Fourier
3.16: Spectral Windows
3.16: Popular Windows
rectangular

Hamming

Hanning

Black man
3.17 ECG Spectral
3.18 EMG Spectral
3.19: EEG Spectral
3.20: Tools for spectrum analysis
 Ch 4: Signal Filtering
• Why filtering?
 4.1 Filter Concept
• Each signal has frequencies range (spectrum)
• Noises is other frequencies and/or same
frequencies range but emitted from other
sources
• The noises from other range (spectrum) can be
eliminated (reduced) with electric circuit and/or
math function.
• Electric/Electronic circuit possible absorb the
noise frequencies
• The math function built from model of electric/
electronic circuit called digital (numerical) filters
 4.1 Physic Filters
• Low-pass filters
Allow low frequencies go through
• High pass filters
Allow high frequencies go through
• Band-pass filters
Allow frequencies in band (range) go through
• Band-stop (Notch) filters
Does not allow frequencies in band (range) go
through
 4.2: Low-pass filters

Thụ động
Tích cực
1
VO ( jw) jwC 1
= = 0 dB
Vi ( jw) R+ 1 1 + jwRC
. Bode
jwC -3 dB

1
x 1/RC ω
0.707 Linear Plot

0 1/RC ω
4.3: High-pass filters
4.4 Order of Filter
Hilbert Problems and Signal Analysis
Relationship
Hilbert - Riemann Problems Original

???

Definition
 Hilbert Transform

Cauchy principal value

Why Cauchy principle value? t=0? Integral(t) = Infinity

Physic phenomenon
Hilbert transform basic functions
Applications
 The topic in Biomedical Engineering?
• EMG analysis
• EEG analysis
• Auscultation analysis
• Ultrasonic imaging
• Voice therapy diagnosis
 Limitations of Fourier Transform
• Limitation of the Fourier Analysis
- For linear input and output
ax1(t)+bx2(t)+cx3(t)=dy1(t)+gy2(t)+hy3(t)
- Stationary and Periodic
• What is the main problems
- Non-stationary
- Local nonlinearity
- Sinusoidal waveform familiar
- Does not match with special waveform delta
pulse like
 Non-stationary wave processing
method
• Spectrogram
• Wavelet analysis
• Wigner-Ville distribution
• Evolutionary spectrum
• Empirical orthogonal function expansion
• Smooth moving average
• Trend least square estimation
 Gabor Transform
+

Short-time Fourier Transform

Gabor Transform

Optimum resolution for frequency and time

2D-Resolution
 Wavelet Transform
+
Wavelet Transform

Morlet also called Gabor wavelet

Optimum Resolution Distribution

Fourier vs Wavelet
Stockwell Transform (S-Transform)
 Limitations
• Fourier provide uniformly resolution in vertical
and/or horizon but conflict between frequency
and time resolution.
• Wavelet provide partial resolution depending on
the order of the decomposition. However the
distribution of the resolution in both time
frequency are not uniform
• Stockwell is special case of Fourier or Wavelet
buts the drawback is the distribution of
resolution.
 Wigner Ville Distribution
• Idea: Overcome the limitation of resolution in
both Fourier and Wavelet
- Because of the window
Function the transforms is
discretizing and delayed
- To remove the delay,
Apply the autocorrelation
function
 Autocorrelation

Definition
Wiener–Khinchin theorem

From Autocorrelation, the distribution of frequency can be built

 Wigner Ville
Shift to τ/2

Resolution go to infinity in continuous domain and limitation in

sampling frequency in discrete domain
Wigner-Ville Example
 Hilbert-Huang Transform (HHT)
• Wigner-Ville is the universal utility for signal
processing. However, it need identified the object to
apply.
• Huang based on Hilbert Transform (a special case of
Wigner-Ville) with Empirical Mode Decomposition
(EMD), extracts the Intrinsic Mode Function (IMF) in
definition of:
The EMD Algorithm
Hilbert transform
HHT Example