I'm trying to write my first signal detection algorithm in Matlab... but it's been several years since I've done any DSP... so it's slow going. if anyone has a little free-time, can you look at these M-files and see if you can spot what I'm doing wrong? the data files can be acquired here: http://www.podoboq.com/podoboq=AD/basic_detect.zip the two examples are simple (Class 1 - everything known a priori) versions of what I am trying to do. Unfortunately, I only have 299 data points to work with, and a sampling rate of 1Hz (Nyquist =3D=3D 0=2E5Hz). I am using a Wiener estimation filter for pre-processing, and for the detection I am using an uncorrelated Wiener-style high-resolution detection filter. "run_me1.m" uses a single sine pulse centered at 100 seconds for the signal I'm trying to detect, and "run_me2.m" has two sine pulses, one centered at 100 seconds and the other centered at 250 seconds. I'm not sure if I'm making a DSP-related error or just an error in the way I'm manipulating the data in Matlab, but it doesn't seem to work. after the estimation phase, you can see the pulse coming out at 100 seconds, but then the final output seems like garbage. Aren't you supposed to get a narrow impulse at each place where a detection occurs??? I understand that there is a stistical aspect to this as wel, but I haven't even got that far yet. What about SNR? In this case the ratio is 1:1. Do I need to RMS-normalize something before filtering? I've experimented with all sorts of things, but maybe I'm just forgetting something dumb! I'd hugely appreciate any help. thnx!!!=20 -ColdCold
who understands signal detection?
Started by ●March 7, 2005
Reply by ●March 7, 20052005-03-07
"theColdest" <uyay6acg-th5x@yahoo.com> wrote in message news:1110237592.957310.168590@o13g2000cwo.googlegroups.com... I'm trying to write my first signal detection algorithm in Matlab... but it's been several years since I've done any DSP... so it's slow going. if anyone has a little free-time, can you look at these M-files and see if you can spot what I'm doing wrong? the data files can be acquired here: http://www.podoboq.com/podoboq�/basic_detect.zip the two examples are simple (Class 1 - everything known a priori) versions of what I am trying to do. Unfortunately, I only have 299 data points to work with, and a sampling rate of 1Hz (Nyquist == 0.5Hz). I am using a Wiener estimation filter for pre-processing, and for the detection I am using an uncorrelated Wiener-style high-resolution detection filter. "run_me1.m" uses a single sine pulse centered at 100 seconds for the signal I'm trying to detect, and "run_me2.m" has two sine pulses, one centered at 100 seconds and the other centered at 250 seconds. I'm not sure if I'm making a DSP-related error or just an error in the way I'm manipulating the data in Matlab, but it doesn't seem to work. after the estimation phase, you can see the pulse coming out at 100 seconds, but then the final output seems like garbage. Aren't you supposed to get a narrow impulse at each place where a detection occurs??? I understand that there is a stistical aspect to this as wel, but I haven't even got that far yet. What about SNR? In this case the ratio is 1:1. Do I need to RMS-normalize something before filtering? I've experimented with all sorts of things, but maybe I'm just forgetting something dumb! I'd hugely appreciate any help. thnx!!! -ColdCold ************** If you really mean that the sample rate == Nyquist then the data isn't sampled at a high enough rate. It is necessary that the sample rate > Nyquist or, said another way, Bandwidth<fs/2 and NOT Bandwidth==fs/2. The link to the files was broken when I tried it.... Fred
Reply by ●March 7, 20052005-03-07
"Fred Marshall" <fmarshallx@remove_the_x.acm.org> wrote in message news:f62dnSe6hZ8ZYrHfRVn-1w@centurytel.net...> > "theColdest" <uyay6acg-th5x@yahoo.com> wrote in message > news:1110237592.957310.168590@o13g2000cwo.googlegroups.com... > > the data files can be acquired here: > http://www.podoboq.com/podoboq�/basic_detect.zip > > > -ColdCold > > ************** > > The link to the files was broken when I tried it....Strange--there is some kind of hidden dash character in the URL right before /basic. This one works: http://www.podoboq.com/podoboq/basic_detect.zip
Reply by ●March 8, 20052005-03-08
where did I say that the sample rate == Nyquist? Please re-read. fs == 1 Hz. meaning Nyquist == 0.5 Hz. data is at http://www.podoboq.com/podoboq/basic_detect.zip thank you, one and all...
Reply by ●March 8, 20052005-03-08
I think you can only expect sharp maxima in the correlation, if the bandwidth is high. If the signal is rather smooth, the maximum will also be. Andre theColdest wrote:> where did I say that the sample rate == Nyquist? Please re-read. > > fs == 1 Hz. meaning Nyquist == 0.5 Hz. > > data is at http://www.podoboq.com/podoboq/basic_detect.zip > > thank you, one and all... >-- Please change no_spam to a.lodwig when replying via email!
Reply by ●March 8, 20052005-03-08
"theColdest" <uyay6acg-th5x@yahoo.com> wrote in message news:1110269468.175891.198290@g14g2000cwa.googlegroups.com...> where did I say that the sample rate == Nyquist? Please re-read. > > fs == 1 Hz. meaning Nyquist == 0.5 Hz. > > data is at http://www.podoboq.com/podoboq/basic_detect.zip >Oh! A nontrivial slip of the keyboard..... sorry about that. if fs ==1Hz then Nyquist is < 0.5Hz and not == 0.5Hz! Or, the other way around, if Nyquist is 0.5Hz then fs must be >1.0Hz. The only difference here being == vs. > or < depending on the expression. Fred
Reply by ●March 8, 20052005-03-08
Reply by ●March 8, 20052005-03-08
hmmmm, the signal in my example is a blackman-windowed sine pulse lasting 20 seconds, with white noise added. is that too smooth a signal to be detected?? gosh, I hope not. it seems like there must be just one simple piece that I am doing wrong? No?
Reply by ●March 9, 20052005-03-09
Fred Marshall wrote: ...> if fs ==1Hz then Nyquist is < 0.5Hz ...I think too much can be made of that truth. That condition assures that the original waveform can be reconstructed. If one has equality instead, the component at Fs/2 can't be completely reconstructed but there is no aliasing. Is not completely being reconstructed a more serious impediment than having been removed? Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●March 9, 20052005-03-09
meanwhile... we digress. ( this has nothing to do with my signal detection problem. :-) if somebody can unzip the data into a Matlab working directory and try them out... shouldn't any detection filter (e.g. a classical detector or a Wiener-uncorrelated high-res detector, etc.) produce a narrow pulse around 100 seconds (and in the second example, also at 250 seconds) ????? I use a Wiener estimation filter to pre-process, but its results look pretty much the same in both examples (a pulse coming out only around 100 seconds) which seems great in example one, but in example two, where is the pulse at 250 seconds. (weird) uggggh.
