Hi, My groups project involves detecting the presence of a train. Our idea is: - send a square wave (one pulse) with a highly skewed duty cycle down the railway track - if a train is present it will create a short circuit which our pulse would reflect off - we would then measure the 'echo' time difference between our sent pulse and the received reflection A sub-system which needs to be looked at is filtering the reflection signal we receive. There will be a lot of noise which is imparted onto the signal from several areas such as the traction return current (probably going to be 50Hz noise in its nicest form) and noise from the trains motors (rectification process). I was wondering what the best options would be, in terms of filtering. As a square wave is made up of a fundamental frequency and then the odd harmonics thereafter, how can I ensure the filter outputs a signal which still represents a square wave? I have to give several suggestions which could be used. I was thinking I would look into passive filters as probably the worst solution (due to component drift etc....), with active filters using op-amps etc being another. But the main solution would be to use dsp, however I have very little knowledge on this subject and therefore I do not know what sort of filter I would be looking for (eg: FIR, IIR) Any input into this would be hugely helpful. Thank you for any help
Filtering options for reflected signal with noise
Started by ●April 15, 2008
Reply by ●April 15, 20082008-04-15
John121 wrote:> Hi, > > My groups project involves detecting the presence of a train. Our idea > is: > - send a square wave (one pulse) with a highly skewed duty cycle down the > railway trackYou need to be clearer. "One pulse" doesn't fit with "duty cycle".\> - if a train is present it will create a short circuit which our pulse > would reflect off > - we would then measure the 'echo' time difference between our sent pulse > and the received reflectionWill it? Are the rails insulated one from the other? What is the characteristic impedance and propagation velocity of the rail pair considered as a two-wire transmission line? Given the spacing, what is the lower usable frequency in that mode?> A sub-system which needs to be looked at is filtering the reflection > signal we receive. There will be a lot of noise which is imparted onto the > signal from several areas such > as the traction return current (probably going to be 50Hz noise in its > nicest form) and noise from the trains motors (rectification process).Why does the reflection need to be filtered? What part of it needs to be removed?> I was wondering what the best options would be, in terms of filtering. > > As a square wave is made up of a fundamental frequency and then the odd > harmonics thereafter, how can I ensure the filter outputs a signal which > still represents a square wave?So, which is it: a square wave, or a single pulse?> I have to give several suggestions which could be used. I was thinking I > would look into passive filters as probably the worst solution (due to > component drift etc....), > with active filters using op-amps etc being another.You haven't said what the filter is for, so suggestions must be vague. Component drift is only a problem if the cut-off frequency must be precise. I don't understand why that should be.> But the main solution would be to use dsp, however I have very little > knowledge on this subject and therefore I do not know what sort of filter I > would be looking for (eg: FIR, IIR) > > Any input into this would be hugely helpful. > > Thank you for any helpJerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●April 15, 20082008-04-15
On Apr 15, 11:44�am, Jerry Avins <j...@ieee.org> wrote:> John121 wrote: > > Hi, > > > My groups project involves detecting the presence of a train. Our idea > > is: > > - send a square wave (one pulse) with a highly skewed duty cycle down the > > railway track > > You need to be clearer. "One pulse" doesn't fit with "duty cycle".\ > > > - if a train is present it will create a short circuit which our pulse > > would reflect off > > - we would then measure the 'echo' time difference between our sent pulse > > and the received reflection > > Will it? Are the rails insulated one from the other? What is the > characteristic impedance and propagation velocity of the rail pair > considered as a two-wire transmission line? Given the spacing, what is > the lower usable frequency in that mode? > > > A sub-system which needs to be looked at is filtering the reflection > > signal we receive. There will be a lot of noise which is imparted onto the > > signal from several areas such > > as the traction return current (probably going to be 50Hz noise in its > > nicest form) and noise from the trains motors (rectification process). > > Why does the reflection need to be filtered? What part of it needs to be > removed? > > > I was wondering what the best options would be, in terms of filtering. > > > As a square wave is made up of a fundamental frequency and then the odd > > harmonics thereafter, �how can I ensure the filter outputs a signal which > > still represents a square wave? > > So, which is it: a square wave, or a single pulse? > > > I have to give several suggestions which could be used. I was thinking I > > would look into passive filters as probably the worst solution (due to > > component drift etc....), > > with active filters using op-amps etc being another. > > You haven't said what the filter is for, so suggestions must be vague. > Component drift is only a problem if the cut-off frequency must be > precise. I don't understand why that should be. > > > But the main solution would be to use dsp, however I have very little > > knowledge on this subject and therefore I do not know what sort of filter I > > would be looking for (eg: FIR, IIR) > > > Any input into this would be hugely helpful. > > > Thank you for any help > > Jerry > -- > Engineering is the art of making what you want from things you can get. > �����������������������������������������������������������������������Jerry, FYI - this guy already admitted on eng-tips.com that this was a student post. I am guessing it will get red flagged and removed PDQ.
Reply by ●April 16, 20082008-04-16
Noway2 wrote: ...> Jerry, > > FYI - this guy already admitted on eng-tips.com that this was a > student post. I am guessing it will get red flagged and removed PDQ.Hi. I assume you mean red flagged and removed from eng-tips.com. As far as I know, that doesn't happen at comp.dsp. I don't mind talking to students. Some of the regulars here were students once. Not doing someone's homework isn't the same as encouraging him to think more clearly. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●April 27, 20082008-04-27
I would have thought it was obv that I'm a student. Whats the harm in asking for guidance if you aren't sure where to go???? Anyway.... The project is as follows: We have decided to measure the input impedance of the section of track (they are insulated at given points). The input impedance is going to change depending on where the train is on the line. We can then use this input impedance to work out the physical location of the train. I am still looking into the filtering of the signal. The reason filtering is going to be needed is because there is all sorts of noise on the line which is going to skew our signal. We are sending a 20V, 1kHz signal down the line. So my thinking behind it is that if we are comparing 2 signals to work out the input impedance of the section then surely if one of the signals is being affected by the nosie then some sort of filter is going to be needed to remove the noise as much as possible. What I was hoping for was some help regarding types of filters which you think my be applicable. Most of the noise is going to be around 47Hz to 52Hz from the traction return. One thing I am unsure about is how would the 3000 amps returing down the line affect our signal? Thanks ps: sorry for being really vague in OP.
Reply by ●April 27, 20082008-04-27
John121 wrote:> I would have thought it was obv that I'm a student. > > Whats the harm in asking for guidance if you aren't sure where to go???? > > Anyway.... > > The project is as follows: We have decided to measure the input impedance > of the section of track (they are insulated at given points). The input > impedance is going to change depending on where the train is on the line.Considering the track as a transmission line, the characteristic impedance and propagation velocity depend only on the geometry of the rails and the material of the ties. The location of the short that you hope the train represents determines the round-trip time of your pulse. The basic principles are used to locate faults in transmission cables. The subject is called "time-domain reflectometry".> We can then use this input impedance to work out the physical location of > the train.Such frequency-domain analysis is harder to do.> I am still looking into the filtering of the signal. The reason filtering > is going to be needed is because there is all sorts of noise on the line > which is going to skew our signal. > We are sending a 20V, 1kHz signal down the line. > > So my thinking behind it is that if we are comparing 2 signals to work out > the input impedance of the section then surely if one of the signals is > being affected by the nosie then some sort of filter is going to be needed > to remove the noise as much as possible.If the probe and power signals have different frequencies, a bandpass filter seems like the obvious way. What is the difficulty?> What I was hoping for was some help regarding types of filters which you > think my be applicable. Most of the noise is going to be around 47Hz to > 52Hz from the traction return.By type, do you mean analog, digital IIR, etc? If it works, it doesn't matter.> One thing I am unsure about is how would the 3000 amps returing down the > line affect our signal?You'll need to find that out. There are other problems: The wavelength of a 1 KHz signal in free space is 30 Km, and more than half that on an open-wire transmission line. Don't expect finer resolution that about a tenth of a wavelength. You need a much higher frequency (narrower pulse). Typically, rails go in two directions. You will have reflections from both directions to deal with. The insulators that isolate sections of rail at the power frequency may be ineffective at the probe frequency. You may not be able to supply enough current to attain 20V from rail to rail. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●April 27, 20082008-04-27
>Considering the track as a transmission line, the characteristic >impedance and propagation velocity depend only on the geometry of the >rails and the material of the ties. The location of the short that you >hope the train represents determines the round-trip time of your pulse. >The basic principles are used to locate faults in transmission cables. >The subject is called "time-domain reflectometry".>The wavelength of a 1 KHz signal in free space is 30 Km, and more than >half that on an open-wire transmission line. Don't expect finer >resolution that about a tenth of a wavelength. You need a much higher >frequency (narrower pulse).These are the reasons we have changed the method of finding the location of the train. After researching the attenuation and velocity of travel of different frequencies it was seen that the original method would not work so well.>If the probe and power signals have different frequencies, a bandpass >filter seems like the obvious way. What is the difficulty?Yes I understand if our signal is in the kHz and the traction noise is lower at 50Hz then all I need is a bandpass or highpass filter. One thing I am unsure of would the following: If we are measuring the input impedance via measuring the voltage at different points (here is a rough idea of how we intend to measure input impedance: http://www.mitedu.freeserve.co.uk/Theory/inzoz.htm) and then seeing the difference in amplitude and phase to calculate the resistance and reactance which we can then use to work out input impedance. As we are sending the signal down the line, it is going to be reflected back towards the source (either via an O/C or S/C). This reflected signal is going to alter the readings at the 2 points. Would this not affect our measurements? If so then I would need to 'filter' the reflection. What types of filters can be used for this? Also the reflection is going to be different depending on S/C or O/C so I cant seem to think of a way of 'filtering' this. When I say filtering for the reflection I don't mean filtering but rather some subtraction or addition.
Reply by ●April 27, 20082008-04-27
John121 wrote:>> Considering the track as a transmission line, the characteristic >> impedance and propagation velocity depend only on the geometry of the >> rails and the material of the ties. The location of the short that you >> hope the train represents determines the round-trip time of your pulse. >> The basic principles are used to locate faults in transmission cables. >> The subject is called "time-domain reflectometry". > >> The wavelength of a 1 KHz signal in free space is 30 Km, and more than >> half that on an open-wire transmission line. Don't expect finer >> resolution that about a tenth of a wavelength. You need a much higher >> frequency (narrower pulse). > > These are the reasons we have changed the method of finding the location > of the train. After researching the attenuation and velocity of travel of > different frequencies it was seen that the original method would not work > so well. > >> If the probe and power signals have different frequencies, a bandpass >> filter seems like the obvious way. What is the difficulty? > > Yes I understand if our signal is in the kHz and the traction noise is > lower at 50Hz then all I need is a bandpass or highpass filter.1 KHz is only the 20th harmonic of 50 Hz, and solid-state controls are rich with harmonics. You would do well do characterize the noise before proceeding further with the design.> One thing I am unsure of would the following: > If we are measuring the input impedance via measuring the voltage at > different points (here is a rough idea of how we intend to measure input > impedance: http://www.mitedu.freeserve.co.uk/Theory/inzoz.htm) and then > seeing the difference in amplitude and phase to calculate the resistance > and reactance which we can then use to work out input impedance.What approximate impedance do you expect? My guess would be milliohms. What mechanism can make th input and output impedances of a passive network differ?> As we are sending the signal down the line, it is going to be reflected > back towards the source (either via an O/C or S/C). This reflected signal > is going to alter the readings at the 2 points. Would this not affect our > measurements?You know what you send, the interesting part is what returns. The reflected signal is your data.> If so then I would need to 'filter' the reflection. What types of filters > can be used for this? Also the reflection is going to be different > depending on S/C or O/C so I cant seem to think of a way of 'filtering' > this. When I say filtering for the reflection I don't mean filtering but > rather some subtraction or addition.Directional couplers can separate outbound and return signals at RF. As far as I know, they are impractical at audio frequencies and unworkable on railroad tracks. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●April 27, 20082008-04-27
Does anyone think a multi-ton engine running on steel wheels on steel tracks does *not* short the rails? They don't mount wheels individually. Two wheels and the axle are one solid piece. It would be nearly impossible for the engine to not short the rails. In fact, nearly all train detectors operate on the shorting of the rails... at least they did some 30 odd years ago. Railroads change about as much in 30 years as other companies do in a month! But I suppose that anyone not familiar with railroads would not know any of this. On Apr 27, 1:42 pm, "John121" <adriane...@hotmail.com> wrote:> >Considering the track as a transmission line, the characteristic > >impedance and propagation velocity depend only on the geometry of the > >rails and the material of the ties. The location of the short that you > >hope the train represents determines the round-trip time of your pulse. > >The basic principles are used to locate faults in transmission cables. > >The subject is called "time-domain reflectometry". > >The wavelength of a 1 KHz signal in free space is 30 Km, and more than > >half that on an open-wire transmission line. Don't expect finer > >resolution that about a tenth of a wavelength. You need a much higher > >frequency (narrower pulse).I am not aware of any reason to consider the frequency rather than the pulse width and edge rate. There are both radars and sonars which send single single pulses and watch for the return. The rep rate is very low and therefore the frequency. But the pulses are short with a fast rise time. This allows the round trip time to be measured very accurately.> These are the reasons we have changed the method of finding the location > of the train. After researching the attenuation and velocity of travel of > different frequencies it was seen that the original method would not work > so well.What sort of distances are you trying to measure? Typically train detectors are measuring distances in the range of a couple of miles or less. For crossings and even for track sections, this is done with a DC current. But this only measures the presence of the train, not the distance. What data do you have for attenuation of signals on rails?> >If the probe and power signals have different frequencies, a bandpass > >filter seems like the obvious way. What is the difficulty? > > Yes I understand if our signal is in the kHz and the traction noise is > lower at 50Hz then all I need is a bandpass or highpass filter. > > One thing I am unsure of would the following: > If we are measuring the input impedance via measuring the voltage at > different points (here is a rough idea of how we intend to measure input > impedance:http://www.mitedu.freeserve.co.uk/Theory/inzoz.htm) and then > seeing the difference in amplitude and phase to calculate the resistance > and reactance which we can then use to work out input impedance. > > As we are sending the signal down the line, it is going to be reflected > back towards the source (either via an O/C or S/C). This reflected signal > is going to alter the readings at the 2 points. Would this not affect our > measurements? > > If so then I would need to 'filter' the reflection. What types of filters > can be used for this? Also the reflection is going to be different > depending on S/C or O/C so I cant seem to think of a way of 'filtering' > this. When I say filtering for the reflection I don't mean filtering but > rather some subtraction or addition.I thought you wanted the reflected signal to alter your measurement? If not, how will the train be detected? If you are trying to measure long distances, you might be able to measure DC resistance. But tracks are often cut and joined with links to allow for expansion and contraction of the rail. The joint is typically bridged with a wire welded the to rails just to allow the detectors to work. If this is always true in your case, then a DC measurement should suffice. To be honest, I don't see how the "impedance" measurement will work or why you need a low frequency AC signal. Do you have an idea of what the impedance of the rails will be for a 1 kHz signal? How do you expect the impedance to vary with the position of the train? One thing that does make this more complex is that you obviously are working with electric trains running off of a powered third rail. One of the rails will be the return path which means you could find some very high potentials on that rail. You would need to do a lot more than just "filter" the incoming signal. I would suggest that you use transformer coupling to connect to the rail for safety reasons.
Reply by ●April 27, 20082008-04-27
>I am not aware of any reason to consider the frequency rather than the >pulse width and edge rate. There are both radars and sonars which >send single single pulses and watch for the return. The rep rate is >very low and therefore the frequency. But the pulses are short with a >fast rise time. This allows the round trip time to be measured very >accurately.Ok. This is what the thinking behind not being able to send a pulse down the line is: We want to send out a single pulse and wait for the reflection. The time taken between sent pulse and returned relection is recorded and with the known velocity of the pulse the distance to the S/C or O/C can be worked out. Now we want to have sections roughly 1/2 mile long, therefore the signal would have to at most travel 1 mile. Here are some figures of velocity of travel and attenuation with frequency: 1kHz: attenuation constant: 0.0011 velocity: 3% c 10kHz: attenuation constant: 0.0037 velocity: 7% c 100kHz: attenuation constant: 0.0104 velocity: 18% c I'm slightly confused because when I say pulse I mean just 1 period of a sqaure wave. And a square wave is made up of a fundamental freq and infinite odd harmonics. If we were to send a 1kHz sqaure wave (1 period) at 20V then it could only get 1 mile before being less than 0.5V. The wavelength of this would also be around 9000m therefore a full period would not even occur along a 1 mile track. A 100kHz, 20V sqaure wave would attenaute below 0.5V after only 200m which makes it useless in terms of this idea. Now am I completely wrong in my thinking of a pulse? All we really need is a spike in voltage for a very short time which would propagate down the line and then return some reflection. Is a pulse with a very small width and fast rise time not made up of very high frequencies? This is what is confusing me?
