Corey Snyder Ben Eng Topics LSIC Systems and BIBO Stability Impulse Response and Convolution ZTransform DTFT and Frequency Response Sampling Discrete Fourier Transform DFT Fast Fourier Transform FFT ID: 788319
Download The PPT/PDF document "HKN ECE 310 Final Exam Review Session" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
HKN ECE 310 Final Exam Review Session
Corey Snyder
Ben Eng
Slide2Topics
LSIC Systems and BIBO Stability
Impulse Response and Convolution
Z-TransformDTFT and Frequency Response
Sampling
Discrete Fourier Transform (DFT)
Fast Fourier Transform (FFT)
Circular Convolution
Digital Filter Design
Rate Conversion and Multirate Systems
D/A Conversion
Slide3LSIC Systems
Linearity
Satisfy Homogeneity and Additivity
Can be summarized by SuperpositionIf
and
then
Shift InvarianceIfthen andCausalityOutput cannot depend on future input values
BIBO Stability
Three ways to check for BIBO Stability:
Pole-Zero Plot (more on this later)
Absolute summability of the impulse responseGiven
then the system is BIBO stable
A bounded input
yields a bounded output Ex: vs. Absolute Summability
Slide5Impulse Response
]
is the impulse response
System output to an
input
Convolution in the time/sample domain is multiplication in the transformed domain, both the z-domain and frequency domain.
Convolution
System must be:
Linear
Shift Invariant
Popularly done graphically
Can also be done algebraically
Z-Transform
We mainly focus on the one-sided, or unilateral, z-transform
Typically perform inverse z-transform by inspection or by Partial Fraction Decomposition
Important properties:
Multiplication by n:
Delay Property #1:
Make sure to note the Region of Convergence (ROC) for your transforms!More in the next slide!DTFT is only defined if the ROC contains the unit circle
Slide8BIBO Stability Revisited
Pole-Zero Plot
For an LSI system: if the ROC contains the unit circle, this system is BIBO stable
The ROC is anything greater than the outermost pole if the system/signal is causal
The ROC is anything less than the innermost pole if the system/signal is anti-causal
If we sum multiple signals, the ROC is the
intersection of each signal’s ROCWhat if the ROC is or ? This is marginally stable, but unstable for ECE 310 purposesFor unstable systems, you are commonly asked to find a bounded input that yields an unbounded output. Few ways to do this:Pick an input that excites the poles of the system.If the system’s impulse response is not absolutely summable, will work frequently works too, like when is unbounded, e.g.
Slide9Discrete Time Fourier Transform
Important Properties:
Periodicity!
Linearity
Symmetries (Magnitude, angle, real part, imaginary part)
Time shift and modulation
Product of signals and convolutionParseval’s RelationKnow your geometric series sums!
Slide10Frequency Response
For any stable LSI system:
What is the physical interpretation of this?
The DTFT is simply the z-transform evaluated along the unit circle!
It makes sense that the system must be stable and LSI since the ROC will contain the unit circle, thus ensuring that the DTFT is well defined
Why is the frequency response nice to use in addition to the z-transform?
is an eigenfunction of LSI systemsBy extension:
Slide11Magnitude and Phase Response
Very similar to ECE 210
Frequency response, and all DTFTs for that matter, are
periodic
Magnitude response is fairly straightforward
Take the magnitude of the frequency response, remembering that
= 1For phase response:Phase is “contained” in termsRemember that cosine and sine introduce sign changes in the phaseLimit your domain from to .For real-valued systems:Magnitude response is even-symmetricPhase response is odd-symmetric
Slide12Ideal A/D Conversion
Sampling via an impulse train will yield infinitely many copies of the analog spectrum in the digital frequency domain
Discrete Fourier Transform
What is the relationship between the DTFT and the DFT?
DFT Properties
Circular
shift
Circular modulationCircular convolutionWe must amend our DTFT properties with the “circular” term because the DFT is defined over a finite length signal and assumes periodic extension of that finite signal.
Slide15Zero-PaddingWe can improve the resolution of the DFT simply by adding zeros to the end of the signal
This doesn’t change the frequency content of the DTFT!
Instead, it increases the number of samples the DFT takes of the DTFT
This can be used to improve spectral resolution
Slide16Fast Fourier Transform
Computational efficient implementation of the DFT
Ordinary DFT requires N
2 multiplies and N(N – 1) addsFFT requires only
computations
Two main forms of FFT
Decimation in TimeDecimation in Frequency (Not covered in this class)Use Butterfly Structures to represent multiplies and adds
Slide17Decimation in Time
Divide sequence into two groups
Before we continue, remember that
We can form the first half of the DFT from these two sequences
And the second half…
Continually halve the sequences until you reach size 2
Fast Linear Convolution
In order to obtain system response, we can multiply DFTs and take inverse DFT
Be careful, this is not the same as convolution in time, but rather cyclic convolution in time
Therefore, in order to perform linear convolution from DFTs, we must first zero pad signals in order to make wrap-around terms go to zeroIf
has length
and
has length , then the resulting convolution is of length .Pad with zerosPad with zerosThis will allow multiplication of FFTs to produce linear convolution result from cyclic convolutionNote: we also typically do extra padding so that the signals are of a length that is a power of the FFT’s radix.
Slide19Linear Phase Filters
FIR vs. IIR – is there feedback?
Types of FIR Filters:
Type 1Odd Length and Even SymmetricType 2Even Length and Even SymmetricType 3Odd Length and Odd SymmetricType 4
Even Length and Odd Symmetric
Slide20Linear Phase Filter Design
Why linear phase?
Satisfy causality and filter should have finite number of terms
Generalized Linear Phase vs. Linear PhaseBegin with ideal magnitude response,
, for a length N filter
Introduce linear phase
Perform inverse DTFT of to obtain Window with window function
Slide21Digital Filter Structure
Suppose we have the LCCDE:
we can express the system in two different
Direct Forms
.
Direct Form IDirect Form II
Slide22Tips for Linear Phase Filter Design
Perform the IDTFT integral for high-pass filters from
.
This will allow you to do one integral instead of two
Only design low-pass filters!
Why?
EasierHow?Use modulation property (multiplication by cosine) to obtain other filters from low-pass filters
Slide23Upsampling
If we upsample by
, we will interpolate
zeros between each sample
if
mod
; and else What happens in the frequency domain?Think about what happens when we oversample a signal, i.e. above NyquistWhat does the frequency response look like after upsampling?Shrink x-axis by factor of What should be in order to obtain a desirable frequency response?Remove extra copies and correct amplitude for conservation of energy
Slide24Downsampling
If we downsample by
, we keep every
sample (decimate the rest)
What happens in the frequency domain?
Frequency response stretches by a factor of
Amplitude reduces by a factor of (think conservation of energy)Anti-aliasing filter prevents downsampling from aliasing our signal
Slide25Ideal D/A
Want
, but we want to only take one copy of the DTFT
Thus, we should low-pass filter from
(domain of the central copy of the DTFT)
Remember that
Remember that multiplication in the frequency domain is convolution in the time domainThus, , where is obtained by multiplying y[n] by an impulse train
Slide26Realizable D/A: Zero-Order Hold
Ideal D/A is not practical because generating delta impulses is not achievable
Zero-Order Hold (ZOH) gives us a suitable approximation to the Ideal D/A
The ZOH multiplies each sample by a rectangular pulse of width T (our sampling rate)Thus,
where
is the rectangular pulse provided by the ZOH
is an analog filter that corrects the distortion presented by the ZOH
Slide27How does the ZOH Change our Spectrum?
Slide28Upsampled D/A
Upsampling prior to D/A conversion can make recovery simpler
i.e. Compensator
can be simpler to implement
Don’t forget that we filter after upsampling!
Full system is then:
[LPF][ZOH]Upsampling effectively increases our sampling frequency, thus our ZOH pulse can be narrower and give us a better staircase approximationThis ‘smoother staircase’ will be easier to rectify with the compensatori.e. the transition bandwidth will be largerIn the frequency domain, we see the frequency axis compress by ; however, the analog frequencies upon recovery do not change!
Slide29P1: CTFT to DTFT
The continuous time Fourier Transform (CTFT) of
is given as:
between
and
andDetermine andforand
Slide30P2 : Non-unique Digital Frequencies
A
cosine is sampled at
. The digital frequency that it is mapped to is
.
Which of the following
sampling frequencies could cause this?a.) b.) c.) c.)
Slide31P3: DFT Matrices and the DFT
Draw the
DFT matrix
Given a length 5 sequence
, and length 8 sequence
,which is just a zero padded
What are the common terms in the DFT of these two sequences?
Slide32P4: DFT Shift Property
Given length 5 sequence
with corresponding transform
. Determine the DFT of
.
P5: Circular Convolution
Determine the circular convolution of sequences
. What is the equivalent matrix operation?
P6 : La Mariposa (The Butterfly)
Slide35Butterfly Structure Exercise
Draw the butterfly structure of a length 8 Decimation in Time FFT for
P7: Zero Padding
Given sequences
. To what length must they be zero-padded to such that their linear convolution is the same as the cyclic convolution?
Suppose there was a radix-3 FFT algorithm; to what length would both of these sequences need to be zero-padded to?
P8: FIR and IIR Systems
Can you guess which of these systems are FIR and IIR?
P9: Filter Structure
Draw the Direct-Form II structure for the following difference equation:
P10 : GLP Filters
Determine the Filter Type for the following systems:
P11 : Window Method
Design a length 34
HPF with a Hamming
Window and cutoff frequency
.
Please also include in your solution a side of fries and a
small soft drink.
Slide41P12 : Up, Down, and Around
is a bandlimited sequence to
It′s Fourier Transform is given as
Given sampling frequency of
Determine
,
The is ideal with cutoff and magnitude 1.
Slide42P13 : ADC /DSP /DAC/WTF
,
.