Hello It is Required to continuously monitor, the received signal from a sensor and detect the presence of the pulse. The amplitude of the pulse remains constant. After A/D conversion, we were trying to pass the discrete time signal through a Long Time Analysis(LTA) Filter and Short Time Analysis(STA) Filter and compare their outputs to detect the pulse (start and end points of the pulse). It is required to know these points as precisely as possible. The SNR of the signal is 5 dB. I am unable to decide on the filter - type,order, time constant. Please comment,tell me if there is a fallacy in the above method. Suggest any references to LTA and STA filter design and selection. Is it more advantageous to do the above process in analog domain? I will be grateful to know about any other better ways of solving this problem(references). Thanks in advance. Parthasaraty
Signal detection problem
Started by ●August 27, 2004
Reply by ●August 27, 20042004-08-27
"Parthasarathy" <parth175@yahoo.co.in> wrote in message news:7f126353.0408270311.4d5cf054@posting.google.com...> Hello > > It is Required to continuously monitor, the received signal from a > sensor and detect the presence of the pulse. The amplitude of the > pulse remains constant. After A/D conversion, we were trying to pass > the discrete time signal through a Long Time Analysis(LTA) Filter and > Short Time Analysis(STA) Filter and compare their outputs to detect > the pulse (start and end points of the pulse). It is required to know > these points as precisely as possible. The SNR of the signal is 5 dB.Why is a proposed method being mixed up with the objective here? How can the SNR always be 5dB? - that doesn't sound very real. What does "the pulse" look like? What does the noise look like? Fred
Reply by ●August 27, 20042004-08-27
Fred Marshall wrote:> "Parthasarathy" <parth175@yahoo.co.in> wrote in message > news:7f126353.0408270311.4d5cf054@posting.google.com... > >>Hello >> >> It is Required to continuously monitor, the received signal from a >>sensor and detect the presence of the pulse. The amplitude of the >>pulse remains constant. After A/D conversion, we were trying to pass >>the discrete time signal through a Long Time Analysis(LTA) Filter and >>Short Time Analysis(STA) Filter and compare their outputs to detect >>the pulse (start and end points of the pulse). It is required to know >>these points as precisely as possible. The SNR of the signal is 5 dB. > > > Why is a proposed method being mixed up with the objective here? > > How can the SNR always be 5dB? - that doesn't sound very real.Problems out of a book are often idealized to hammer on a particular point. ... Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●August 27, 20042004-08-27
"Jerry Avins" <jya@ieee.org> wrote in message news:412f5e4e$0$19713$61fed72c@news.rcn.com...> > Problems out of a book are often idealized to hammer on a particular > point.Book? What book? .....
Reply by ●August 27, 20042004-08-27
Fred Marshall wrote:> "Jerry Avins" <jya@ieee.org> wrote in message > news:412f5e4e$0$19713$61fed72c@news.rcn.com... > >>Problems out of a book are often idealized to hammer on a particular >>point. > > > > Book? What book? .....I assumed a text. Given the season. self study, rather than homework. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●August 27, 20042004-08-27
Sounds like a DC shift in stationary white Gaussian noise. I think for a long term integrator and a short term integrator, the F-Ratio test is appropriate with the number of degrees of freedom equal to the the number of samples in the filters. A sliding boxcar filter would be nice from an analysis perspective. There is a book on this subject but I'm drawing a blank on the title. The random telegraph signal from Papoulous (?) might be the source of the problem. In article <412f77f9$0$19708$61fed72c@news.rcn.com>, Jerry Avins <jya@ieee.org> wrote:>Fred Marshall wrote: > >> "Jerry Avins" <jya@ieee.org> wrote in message >> news:412f5e4e$0$19713$61fed72c@news.rcn.com... >> >>>Problems out of a book are often idealized to hammer on a particular >>>point. >> >> >> >> Book? What book? ..... > >I assumed a text. Given the season. self study, rather than homework. > >Jerry
Reply by ●August 28, 20042004-08-28
george.w.bush@whitehouse.com (George Bush) wrote in message news:<SBPXc.18184$Qa4.6770@twister.socal.rr.com>...> Sounds like a DC shift in stationary white Gaussian noise. I think for a long > term integrator and a short term integrator, the F-Ratio test is appropriate > with the number of degrees of freedom equal to the the number of samples in > the filters. A sliding boxcar filter would be nice from an analysis > perspective. There is a book on this subject but I'm drawing a blank on the > title. > > The random telegraph signal from Papoulous (?) might be the source of the > problem. > > In article <412f77f9$0$19708$61fed72c@news.rcn.com>, Jerry Avins > <jya@ieee.org> wrote: > >Fred Marshall wrote: > > > >> "Jerry Avins" <jya@ieee.org> wrote in message > >> news:412f5e4e$0$19713$61fed72c@news.rcn.com... > >> > >>>Problems out of a book are often idealized to hammer on a particular > >>>point. > >> > >> > >> > >> Book? What book? ..... > > > >I assumed a text. Given the season. self study, rather than homework. > > > >JerryAs an alternative to the sliding boxcar, consider using a "leaky integrator" to establish the background noise level: y[n] = (1-alpha)*x[n] + alpha*y[n-1], 0 < alpha < 1 John
Reply by ●August 28, 20042004-08-28
On Fri, 27 Aug 2004 12:16:14 -0400, Jerry Avins <jya@ieee.org> wrote:>Fred Marshall wrote: > >> "Parthasarathy" <parth175@yahoo.co.in> wrote in message >> news:7f126353.0408270311.4d5cf054@posting.google.com... >> >>>Hello >>> >>> It is Required to continuously monitor, the received signal from a >>>sensor and detect the presence of the pulse. The amplitude of the >>>pulse remains constant. After A/D conversion, we were trying to pass >>>the discrete time signal through a Long Time Analysis(LTA) Filter and >>>Short Time Analysis(STA) Filter and compare their outputs to detect >>>the pulse (start and end points of the pulse). It is required to know >>>these points as precisely as possible. The SNR of the signal is 5 dB. >> >> >> Why is a proposed method being mixed up with the objective here? >> >> How can the SNR always be 5dB? - that doesn't sound very real. > >Problems out of a book are often idealized to hammer on a particular >point. >Hi Jerry, This sounded much like a homework problem to me too. I suggest Parthasaraty look into "matched filters". [-Rick-]
Reply by ●August 30, 20042004-08-30
Hello> > Why is a proposed method being mixed up with the objective here?What are you trying to convey?> How can the SNR always be 5dB? - that doesn't sound very real.I intended to say that, the SNR is >= 5dB.> What does "the pulse" look like?The Pulse is a chirp signal.> What does the noise look like?I am not sure, how it looks like, I have assumed to be gaussian. Also, please tell about the various signal detection methods Thanks in advance Parthasarathy
Reply by ●August 30, 20042004-08-30
"Parthasarathy" <parth175@yahoo.co.in> wrote in message news:7f126353.0408300059.162c0f15@posting.google.com...> Hello > > > > > Why is a proposed method being mixed up with the objective here? > > What are you trying to convey? > > > > How can the SNR always be 5dB? - that doesn't sound very real. > > I intended to say that, the SNR is >= 5dB. > > > What does "the pulse" look like? > > The Pulse is a chirp signal. > > > What does the noise look like? > > I am not sure, how it looks like, I have assumed to be gaussian. > > Also, please tell about the various signal detection methods > > Thanks in advance > Parthasarathy> > Why is a proposed method being mixed up with the objective here? > > What are you trying to convey?You said: "we were trying to pass the discrete time signal through a Long Time Analysis(LTA) Filter and Short Time Analysis(STA) Filter and compare their outputs to detect the pulse (start and end points of the pulse)" and, that is a "method". So, I was asking why the proposed method was part of the statement of the objective. That's all. Maybe you have a constraint that requires this approach / half developed, etc. You might consider a matched filter - one that's matched to the chirp. In the time domain, this would amount to correlating the chirp with the input signal - multiply and add/integrate the products. This could be done at each sample interval but perhaps doesn't need to be that frequent. You would be looking for a peak in this result. In the frequency domain you would first zero-pad the chirp and FFT, keeping this as a reference. Then, you would take longer segments of the signal (equal to the length of the padded chirp) and FFT them - perhaps overlapped. Then you would multiply the filter by the signal in the frequency domain and IFFT the result. You would be looking for a peak in this result. This process would identify the presence of the pulse in noise but might not be so good for finding the edges of the pulse. You might use it to find the center of the pulse and measure using some other technique around the center looking for the edges. Maybe that's where long and short term "analysis" comes in. The problem with finding the edges is that it implies a much higher bandwidth. So, the matched filter isn't going to be optimum for that purpose. If you know the frequencies of the chirp, maybe you can match filter over a shorter time period at the leading and trailing edge frequencies - at quite a loss in processing gain. Fred Fred
