Hi! I have to design a pulse time filter. Which means It has to get rid of the pulses which have a time duration of less than 2ms and take >5ms pulses in to account. How can I design a one? I got some clues but couldn't understand. Can anybody explain that, Pls?

{input from band pass filter}---> {Amplifier,Rectifier,LPFilter}---> {SCHMITT trigger}--->{Set-Reset multi vibrator}---> {Buffer stage}--> Output

I got the above schematic from a research paper. But I am not sure about this, because considering the characs of each components, it doesn't make any sense. So give me some ideas please.

thank you!

Bavan

 

That seems like a terribly difficult way to do it. I'd just use a sharp low pass filter to filter out anything above 1/(5ms), and then condition the resulting 5ms pulse back to looking like a square pulse since your filter will wipe out some of the spectral content of the 5 ms pulse. The "hard" part is getting the 2ms pulse taken care of, but a nice sharp LPF will work nicely.

Edit:

I just did the problem in matlab with an 254th order FIR filter and what equates to a schmidt trigger. My resulting waveform had a timewith of 5.08 ms. I hope thats within spec.

 

Thank you! So the schematic is ok. And you mean that first LPF is not required?
What is the use of SR flipflop? Do we actually need a buffer? (the out put is given to a Counter pulse generator)

Bavan

 

I didn't look at the paper/schematic. What I said if you read my post is that the way you describe doing it is far too complicated, and I suggested using a low pass filter and a schmitt trigger. I simulated the idea in matlab and I got rid of the 2ms pulse by widening the 5ms pulse by 80 microseconds. If thats not with in the tolerance, you could sharpen the roll off on the filter, I put the stopband at 300 hz, it could be lessened, but the filter would get longer.

The new schematic would be LPF > schmitt trigger, everything else is unnecessary. Essentially what this idea does is filters out the main lobe of the sinc function created by the 5ms pulse, and then reconditions the signal because the filter also wiped out the tails of the sinc function for the 5ms pulse (and we need those to have a pulse), so I used the schmitt trigger to put them back in. The consequence of doing it this way is you slightly narrow the main lobe of the sinc function (that means widening the pulse) by 80 microseconds. Thats a small price to pay I think.

 

Thanks. I'll do simulations and check. Thanks again.


But I have one more questions. Normally how do you increase the reliability in your systems? My supervisor always blames me that my system has low reliability. I don't know how to do that.

I am designing a system for railway crossing signaling system. My design is

{train detection using a induction method(done)}--->{selective amplifier & filter(done)}---->{pulse time filter(thank you for your design)}--->{counter pulse generation}--->{counter and counter supervision}---->{Atmel controller(done)}---->{RF modules: link(done)} --->{Atmel main controller}-->{track occupied signaling system}

Here I'm planning about having two layer of each so that even one channel is failed other can run(diversity). But supervisor seems not ot be satisfied with it. Wat can I do? Any suggestions?

Thank you!

 

I'm not sure what you mean by reliability.

 

This system is a safety critical system. It should not fail.

 

i wonder... for ur requirement )train monitoring), the system could have been designed in a much more simpler manner....

and abt the pulse time filter... its like u r passing all frequencies greater than 200Hz.... wont a simple high pass filter do the job?

 

The pulse descrimination conceivably could also be done digitally. When a pulse is detected it triggers a one-shot to generate a 2ms pulse. This is compared with the incoming pulse. If the incoming pulse is shorter than the one-shot signal it is ignored, otherwise it is accepted.

Regarding reliability, for critical requirements three circuits are often used with a majority rule. That is, the outputs of at least two of the three must agree. If you have only two circuits it may not always be possible to determine which one failed.

 

ya.... actually doing it in digital domain makes the job musch easier too.... u can use a microcontroller like the PIC also to discriminate between the 2ms and the 5ms pulses...

 

hi Raviram,thank you for your reply.
All other systems have low reliability. A project has already been done by an American company and the result is attached with this reply.

High pass filter will give us high frequency components of both pulses. So I couldn't get your point.

There was a research paper available on IEEE on Axle counting systems titled "An Axle counting system for Automatic Signaling of Trains" . You can read that if you like.

I want some common techniques you all use while developing a system such as having diversity/redundancy. Please, some help over here.

Thank you again.

Bavan

Attached Images

statistical data.JPG (101.7 KB, 9 views)

 

Thanks Crutschow! I have suggested that diversity plan to my supervisor. I'll try to convince him about this. And do you know how to increase the reliability of an RF link? Having a dual line will cost much. So what can I do?

 

A simple 'fail safe' method would be to continually check for data over the RF link, so send one type of data stream for a 'non-event' (no train) and another type for an 'event' (train detected) - if the RF link goes down no data at all will be received, and the system can generate an error condition accordingly.

__________________
PIC programmer software, and PIC Tutorials at:
http://www.winpicprog.co.uk

 

Thats what we suggested. (using extremely different bit streams for detection and no detection ). is that enough for a reliable system? Anyway it seems no other methods are available. So I concentrate on that.

Then, I bought a rf module for testing purpose (rrfq2-433MHz). But there is lot of noise(anonymous bits are received) at the receiver while I tried to send some data. Even when the transmitter is off there are some thing received(no 433MHz tx at lab environment). I can see only some sudden change at the received bit while I 'on' the transmitter but not the one I send. what do you think? Might be some problem with antenna matching? I just used monopole(lambda/2 wire) antenna. Any suggestions? ( program part is ok.)

 

You need to use Manchester coding (or similar), you can't just send standard serial data via an RF link. Check my PIC tutorials, which give an example of how to do it with PIC's.

__________________
PIC programmer software, and PIC Tutorials at:
http://www.winpicprog.co.uk