Interview Questions on DSP - Open Thread - Please Contribute

[ashuisgolu]

dsp interview questions

Hello,


what is the difference between DFT and DTFT?

DTFT ie discrete time fourier transform gives the fourier transform of a discrtete time signal and is continous and periodic in frequency domain. Whereas DFT ie discrete fourier transform also gives samples of the fourier transform of a discrete time signal. It is discrete and periodic in the frequency domain. We will get dft by sampling the DTFT. DFT is more suitable for discrete processing or processing in a DSP processor.

 

[hamran]

dsp questions for interview

"wut is decimation?"

Added after 2 minutes:

"wut is de-interlacing?"
"how could it be useful in prctice?"

 

[chin]

interview questions and answers + dsp

1.What are the Differences between DTFT and DFT
2.What are LTI systems, Linearity theorem, and stability of a system?
3.Spectral efficient and power efficient digital modulation techniques

 

[electricpete]

dsp processors interview questions

Does "zero padding" improve FFT resolution?

Answer: No, because resolution is determined by the length of
observation interval, and zero padding does not increase this
length

Refer
http://home.eng.iastate.edu/~julied/classes/ee524/LectureNotes/l5.pdf

I did see on page 8/20 where he said it does not improve the resolution.

I will have to respectfully disagree. Zero-padding does indeed improve the frequency resolution. The DFT/FFT are "samples" of the DTFT at discrete frequencies. We get more samples (higher resolution) to the true DTFT by zero-padding. There are limitations which cannot be overcome by zero-padding, and you might put these in the category of accuracy, discrimination (of two closely spaced peaks), noise immunity, etc. But there is no doubt in my mind that zero-padding can improve the frequency resolution and the ability to separate peaks.

Page 8 of 50 at the top shows a rectangular window with zero-padding. The zero-padding does NOT change the blue envelope, but it does change the red samples of that envelope. When we time-multiply a sinusoid by the singal, we get frequency convolution. What we end up with is a scaled replica of the window DFT centered on the frequency of the sinusoid. (And if we have two sinusoids, then we end up with two scaled replica's which may overlap). The overall blue shape (representative of leakage) does not change. But the red samples become close together as we zero pad which certainly should improve our ability to see what is going on.

Note that zero-padding does not add any new information. We could in theory reconstruct/interpolate values of the DTFT between bins from the unpadded DFT a number of of other ways (which are failrly computationaly intensive).

===================================

http://home.comcast.net/~electricpete/edaboard/ZeroPadDemoR2.doc

At the above link, I have prepared a Matlab demo that I believe shows an improvement in resolution associated with zero-padding.

There are two closely spaced sinusoids at frequencies f1 and f2:
f1 = 8.96*binwidth = 8.96 hz
f2 = 10.27*binwidth = 10.27 hz

These can clearly be distinguished in the zero-padded spectral plots 4 and 8, but not in the original non-padded spectral plots 2 and 6.

Easier direct comparision is shown on plots 9 and 10:
Plot 9 – superimposes plots 2/4 (no window with/without padding)
Plot 10 – superimposes plots 6/8 (Hanning window with/without padding)

If sure looks like an improvement in resolution to me. The leakage is not eliminated or reduced, but we have more detail (more samples along the DTFT curve) to discern what the underlying pattern is.

One thing that didn't quite make sense to me as I was preparing the writeup is why the apparent frequency of the two peaks is pushed farther apart than actual in the WINDOWED zero-padded spectrum (plot 8_) as compared to the NON-windowed zero-padded spectrum (plot 4) as I discussed in the "overview" section of my linked document. After thinking awhile, I think the answer might be buried in the phase relationships which are not shown on the plots.

 

[electricpete]

interview questions for digital signal processing

An even better demonstration is looking at a single frequency as in the following link::
**broken link removed**


A known 10.3hz signal with white noise is input. Bindwidth is 1hz. FFT predicts frequency as 10hz. Zero padding predicts as 10.3hz. In absence of noise, accuracy of frequency prediction (for single frequency) through zero padding is limited only by numerical accuracy. With noise there is some error introduced, but bias should be 0.

Again my conclusion is that zero-padding does improve resolution. It does not reduce leakage (as can be done by increasing sample length), but it provides a much more detailed picture of the information available from original sample.

 

[coros]

dsp rho interview questions

But in your paper you have also increased the number of samples by adding zeros. The FFT is no longer 1024 point but increased by the number of the padded zeros. Naturally a 8192 point FFT will have better resolution than 1024, not because of the additional zeros. In your paper we can also observe that an abrupt end of the sinusoid produces ripples in frequency domain. That's why we use window functions when padding with zeros.

 

[electricpete]

dsp processor interview questions

Yes, I have added zero's to the end of the windowed sample and taken the FFT of the whole thing. It's called zero-padding of the original signal. If you agree it improves resolution, we are in violent agreement. If you say it is "obvious", I wouldn't disagree. But there was someone who said zero-padding doesn't improve resolution which sparked my comment.

If there was some fellow I missed in this thread who said you can take away data and replace it with zero's to improve resolution (which is not zero-paddding by the way), then I'd like to meet him to tell him he is wrong. That fellow certainly isn't me.

 

[coros]

dsp intervie questions

Then we are in agreement, but we understood the "zero padding" thing differently. In my previous post I wanted to add, to what was already stated:

chegu.balaji wrote:

chickoos wrote:
Does "zero padding" improve FFT resolution?



Answer: No, because resolution is determined by the length of
observation interval, and zero padding does not increase this
length

 

[vinod_261]

objective question in dsp

What are lapped transforms and why they are used.

 

[naresh850]

objective question in digital signal processing

Wat is the difference between circular convolution and linear convolution?
where each one is used?

 

[calm]

sample questions for dsp

who known MRI?

 

[naresh850]

signal processing aptitude question paper

What are the difference between floating point and fixed point processer?

What are the constraint of fixed point processer?

 

[xulfee]

dsp + interview questions

u must've knowledge of matlab programming,it can also be asked

 

[eng_shady00]

dsp interview questions list

u must've knowledge of matlab programming,it can also be asked

Please i want some help in using fdatool in matlab. I designed my filter but don't know how to apply it to my signal.

I do export the filter as an object or to workspace or as an SOS matrix, or even as an m-file. But i can't apply this filter to my original signal.
How can i get that.

Or if u suggest any other method for me to design a BPF and apply it to my signal.
I've really got confused.
Thanks in advance.

 

[Communications_Engineer]

typical interview questions for dsp

Where to use Laplace, Z, Fourier and Hilbert Transforms?

-------------------------------------------------------
Important, right?
-------------------------------------------------------

Use Laplace to find the stability, causality of continuous time systems.

Use Z for same analysis of a discrete time system or for those systems where you have integer delays (result) of fixed sampling. Some people say Z-transform isn't a transform at all, it is a mapping from time to complimentary time (with complex mapping). They say similar stuff about Laplace

Fourier is used for frequency domain analysis (phase and mgnitude response)

Hilbert is used to find, I think Phase-time analysis

Correct me if I'm wrong

 

[Aya2002]

interview questions on signal and system

why we put guard period between symbols and what is the signal will be sent in this period?

 

[xie.qiang]

interview questions on digital signal processing

Why are we using windows when doing FFT?

Why are we using overlap when doing FFT?

What is the trade-off of selecting a window to do FFT?

What is the relationship between phase noise in Frequency and jitter in time domain?

What is the concern when consider your fix-point algorithim when implementing?

..etc...

 

[Communications_Engineer]

how to interview for a dsp position

It WOULD be nice if SOME people can also give answers. Asking questions is easy but answering them is difficult. I hope some people know the answers to their question and should share with us

 

[AdvaRes]

dsp related interview questions

Source:

https://vlsifaq.blogspot.com/search/label/Basic Microelectronics


What is impulse response?
Explain the advantages and disadvantages of FIR filters compared to IIR counterparts.
What is CMRR?
Explain half-duplex and full-duplex communication?
Which range of signals is used for terrestrial transmission?
Why is there need for modulation?
Which type of modulation is used in TV transmission?
Why we use vestigial side band (VSB-C3F) transmission for picture?
When transmitting digital signals is it necessary to transmit some harmonics in addition to fundamental frequency?
For asynchronous transmission, is it necessary to supply some synchronizing pulses additionally or to supply or to supply start and stop bit?
BPFSK is more efficient than BFSK in presence of noise. Why?
What is meant by pre-emphasis and de-emphasis?
Explain 3 dB cutoff frequency? Why is it 3 dB, not 1 dB?
Explain ASCII, EBCDIC?

 

[tibm]

short questions on digital signal processing

@ lena,
Adding to ur answer abt cross correlation and auto correlartion..
Cross correlation determines the similarity b/w 2 processes and auto correlation finds it for the same process but may b different instants ..

Added after 12 minutes:

Source:

https://vlsifaq.blogspot.com/search/label/Basic Microelectronics


What is impulse response?
Explain the advantages and disadvantages of FIR filters compared to IIR counterparts.
What is CMRR?
Why is there need for modulation?
Explain ASCII, EBCDIC?

the response of a steady state system to an impulse,i.e, +vcc for a infitesimally short interval is called the impulse response

FIR hav linear phase as compared to IIR, who hav nonlinear
CMRR - Common Mode Rejection Ratio

Modulation is needed for ensuring less distortion of the baseband signal due to noise abd also to use small antennas for receiving de signal