hello, im currently doing my masters.. i need to decide my topic of my thesis in DSP on non uniform sampling of signals.. your suggestions will be of great help... thanx
need help in non uniform sampling
Started by ●November 16, 2011
Reply by ●November 16, 20112011-11-16
On 11/16/2011 1:07 PM, ankit wrote:> hello, > > im currently doing my masters.. > i need to decide my topic of my thesis in DSP on nonuniform sampling > of signals.. > your suggestions will be of great help... > > thanxI hope in your Masters program you learn to express yourself more clearly. Do you need to select a topic for your thesis that *is definitely going to be* about non-uniform sampling? or Are you asking if a topic dealing with nonuniform sampling is a good idea? (I'm not sure it's a good idea as much work has already been published. But maybe I have too stringent a filter in topic selection.) What is it that *you* want to do? What is it that your professor wants you to do? etc. Fred
Reply by ●November 17, 20112011-11-17
> On 11/16/2011 1:07 PM, ankit wrote: > i need to decide my topic of my thesis in DSP on non uniform sampling > of signals.You need to ask better questions. There is a ton of material you have to read. Learn the applications; learn the algorithms. You can view quite a few of them by searching for "mri algorithms and fft" (364,000 hits). Here's one (449Kb - pdf): http://www.waset.org/journals/waset/v42/v42-34.pdf Then you might search for: NUFFT (yes, it has a name - Non-Uniform FFT). Here's a couple refs: http://math.nyu.edu/faculty/greengar/glee_nufft_sirev.pdf http://www.cscamm.umd.edu/programs/fam04/qing_liu_fam04.pdf You might also spend a few weeks (or months) visiting a university with a good selection of medical imaging journals. You might start with the earliest papers and learn in a historical sequence as to what was discovered and when. Review or tutorial papers can be helpful to cut through the fog. If you can, seek out a researcher at GE, Philips, Siemens, Toshiba or some other manufacturer of medical diagnostic equipment. Most of all, your field of study should be something that fascinates you - something you truly want to learn about. Kevin McGee
Reply by ●November 17, 20112011-11-17
On Nov 16, 5:48�pm, Fred Marshall <fmarshallxremove_th...@acm.org> wrote:> On 11/16/2011 1:07 PM, ankit wrote: > > > hello, > > > im currently doing my masters.. > > i need to decide my topic of my thesis in DSP on nonuniform sampling > > of signals.. > > your suggestions will be of great help... > > > thanx > > I hope in your Masters program you learn to express yourself more clearly. > > Do you need to select a topic for your thesis that *is definitely going > to be* about non-uniform sampling? > > or > > Are you asking if a topic dealing with nonuniform sampling is a good idea? > (I'm not sure it's a good idea as much work has already been published. > � But maybe I have too stringent a filter in topic selection.) > > What is it that *you* want to do? > What is it that your professor wants you to do? > etc. > > FredHi Fred, My proffesor doesnt have a clear idea about the topic. Right now he just asked me to find an application related to non uniform sampling. But i was a bit confused as pointed out by you that most of the work has been published. so i was looking for some help. thanks for the quick reply. Ankit
Reply by ●November 17, 20112011-11-17
On 11/17/2011 10:39 AM, ankit wrote:> On Nov 16, 5:48 pm, Fred Marshall<fmarshallxremove_th...@acm.org> > wrote: >> On 11/16/2011 1:07 PM, ankit wrote: >> >>> hello, >> >>> im currently doing my masters.. >>> i need to decide my topic of my thesis in DSP on nonuniform sampling >>> of signals.. >>> your suggestions will be of great help... >> >>> thanx >> >> I hope in your Masters program you learn to express yourself more clearly. >> >> Do you need to select a topic for your thesis that *is definitely going >> to be* about non-uniform sampling? >> >> or >> >> Are you asking if a topic dealing with nonuniform sampling is a good idea? >> (I'm not sure it's a good idea as much work has already been published. >> But maybe I have too stringent a filter in topic selection.) >> >> What is it that *you* want to do? >> What is it that your professor wants you to do? >> etc. >> >> Fred > > Hi Fred, > > My proffesor doesnt have a clear idea about the topic. > Right now he just asked me to find an application related to non > uniform sampling. > But i was a bit confused as pointed out by you that most of the work > has been published. > so i was looking for some help. > > thanks for the quick reply. > AnkitAnkit, Some thoughts: 1) Tell the professor what you've learned so far. Be informative so you can demonstrate what your research reveals. Use some examples to prove you aren't just blathering. Maybe the professor will give you some ideas that expand on the initial assignment. Or, maybe the professor will decide you could or should pursue something else. 2) If you were doing this without guidance, what would you be working on? You might hint that this seems an interesting area to pursue. 3) Ask some people in the field, perhaps right here in comp.dsp, "What nonuniform sampling issues remain unsolved, poorly examined or illuminated, interesting to the community? ... that would be maybe reasonable for a Master's Thesis. Here's a hint from my experience: Given: - A data stream is collected. The numbers are used to decide whether there is excessive .... whatever. - It is known that the sample rate is too low to satisfy the Nyquist criterion. OR, the sample interval is nonuniform and the Nyquist criterion may or may not be satisfied. - Long term averages are more important than short-term excursions. Required: - What processing might you do on the collected data to determine: 1) The magnitude and lengths of likely short-term excursions above the threshold? 2) A measure of the "effective" data values. Possible approach: Assume that the samples are members of a population rather than attaching any particular fine temporal or frequency measures to the data at all. Calculate the distribution of the values of the data points. What does this tell you? Is that information useful? How might this be applied in situations where signal processing experts have torn their hair out because of the sample intervals involved and their "in the box" thinking? Advanced: What is the relationship between the distribution and the likely spectral character (or some other measure) of the data? Hint: Use the same analysis on data that is adequately and uniformly sampled. Compare with your starting data set or generat a nonunformly sampled set out of the same underlying set. Can any conclusions be reached? Is that useful? Question: Data is known to contain "spikes" of some average rate of occurrence and of some average duration. What is the probability of capturing the peak of a spike given your selected nonuniform sampling method? Question: When is nonuniform sampling *better* than uniform sampling? When is sampling that's "out of phase" with the suspected/known frequencies in the data better than uniform sampling intervals that might be correlated with the data? (Hint: see the distributions above. Might the tails of the distribution be captured?) Now, I think these are good questions at the Masters level because *I* don't know the answers. I've thought about it some but not to the extent that's suggested here. Maybe somebody does and will spoil your fun. Good luck! Fred
Reply by ●November 18, 20112011-11-18
On Nov 16, 4:07�pm, ankit <avsha...@gmail.com> wrote:> hello, > > im currently doing my masters.. > i need to decide my topic of my thesis in DSP on non uniform sampling > of signals.. > your suggestions will be of great help... > > thanxHello, Fred et al, gave some good advice. Think about applications where the data by its very nature tends to be nonuniformly sampled yet some information from that data is desired. I happen to be working with astronomical data (observations and measurements) that will almost always end up being taken nonuniformly. I don't get to pick the weather or times that I have access to the big telescopes! There is a lot written on some general techniques, but you may find a refinement of a technique for specific data is both useful and needed. Most of us who use DSP (its methods and theory) in our work tend to have it applied to very special problems, so maybe if there is a special problem you are interest in and this could be something to pursue. But this is where a good advisor comes in handy, since he is the one to guide you on your work and if you choose something he doesn't know much about, it kind of limits what help he can offer to you. So you need to choose something you are interested in, something your advisor can help you with and, most importantly, something you can be successful with. My 2 cents worth Clay
