Menu
What's new
Find's Treasure Forums

Welcome to Find's Treasure Forums, Guests!

You are viewing this forums as a guest which limits you to read only status.

Only registered members may post stories, questions, classifieds, reply to other posts, contact other members using built in messaging and use many other features found on these forums.

Why not register and join us today? It's free! (We don't share your email addresses with anyone.) We keep email addresses of our users to protect them and others from bad people posting things they shouldn't.

Click here to register!

Need Support Help?

Cannot log in?, click here to have new password emailed to you

PI Filtering

A

Anonymous

Guest
Just found this site. I was reading previous messages about filtering. I have used a comb filter in the past to filter the O/P of the coil-amp. make up a one I/P - 32 O/P, analog mux. 10nF or 100nF capacitors go from each O/P to ground (32 caps)Connect the mux input to the wiper of a pot. This connection is the filters O/P (buffer with an op-amp and LPF the amps O/P with an RC). one side of the pot is the I/P from your coil amp. Generate a fX32 clock with a PLL locked to the pulse frequency to drive the MUX. The pot's R sets the Q. This filter passes ANY periodic waveform as it passes harmonics. It also has a zero degree phase shift. It is tuned by the clock fX32). My version used a 5 bit counter to address four 1 of 8 analog MUX chips. The counter was clocked by the PLL,VCO of a 74HC4046. The counters f/32 O/P went to the phase detector where it was compared to the flyback pulse. A Hi-Z op amp can be connected to the appropriate cap to provide the signal sample. Watch out for the pots R value as this filter can easily provide a massive Q. Start with a low R setting and go up. Too high a Q will cause a slow response to a target. Set the Q for optimum response speed. A 16 position MUX driven with fX16 also works well. Watch your sensitivity go up and noise dissapear when you use this method! You will be able to add more amplification. Use good caps and try to use low on resistance MUX chips, Dave. * * *
 
Hi Dave
Very good description, but I get lost somwhere could you send me the schematic by email I will be very thankful.
Mark
 
Hi Dave,
Sounds interesting. Any chance of posting a schematic?
Eric.
 
OK guys,
I sent an e-mail to both Mark and Eric. I included a partial schematic and and an explanation as to the filters operation. Feel free to post the schematic (I had trouble trying) and a better description than my own, Dave. * * *
 
Hi Dave,
Many thanks for all the information on the comb filter. It is not something that I have used to date and I am still trying to get my head round the principles. As I understand it, the received waveform that is in phase with the sampling, passes through the filter unscathed. Any non-coherent waveform from pickup on the coil is filtered out. Does this also apply to random noise? This is the bane of industrial detectors where you have impulse noise, switching transients, etc.
Eric.
 
I would be most appreciative if someone would post or email me the schematic.
Thanks,
Charles
 
Eric, Think of the multiplexer as a rotary single pole 32 throw switch running in circles. The signal comes in through a resistor to the switch pole. The switch goes all the way around 1000 times a second. each cap is in circuit one at a time. OK, imagine a 1KHz sine wave. For 32 caps then each cap is in circuit for 11.25 degrees of the 360 degrees of one cycle. Each cap now builds up a charge through the resistor each time it comes into circuit. you now have 32 caps each holding a voltage sample at each 11.25 degrees of the sine wave. Only a signal which is synchronous to the clock will be maintained. All others will be non synchronous and will cancel out. Imagine what 50Hz mains hum will do. The capacitors for the 1KHz filter will go around 20 times for one 50 Hz cycle. The caps will not build a charge which is synchronous and the signal will cancel. The operation is much like a synchronous demodulator except that you have 32 of them. The filter will pass the harmonics of the input signal. Due to this, you can pass any repeating waveform that you wish. Sine, square, triangle, saw, or the response curve of a PI receiver. Any non synchronous signal will be severely attenuated by a comb filter. This includes impulse noise, mains hum, auto ignition noise, or a marching band for that matter. The voltage on each capacitor takes a number of switch rotations to build up due to the input resistor. It also takes a while to run down. The resistor therefore sets the Q. The filter will ring on a sinewave when the Q is high just like any other high Q filter. The filter can be made any length. Analog MUX chips are most commonly found in a 1 to 8 configuration. four of these chips will make a 32 position filter. Consider a PI where the pulse repetition frequency is 1KHz. For a 32 position filter, each capacitor is in circuit for 31.25uS. This is equal to about the average sample gate width of a PI. You therefore get the signal resolution required. The Q of the filter is pi * fo * N * R * C. Where fo is the clock/N and N is the number of capacitors switching. R is the input / output resistor. So for a 1Khz filter with for example a 20K resistance and 0.1uF capacitors, using just 8 capacitors we have 3.142 X 1000 X 8 X 20000 * 1 to the minus 7 we get a Q = 50.272. For a 32 position filter and a 32KHz clock and with all else the same we get a Q of 201.088. Remember what I said about the Q going up a lot higher than you want it too. Start with a Q of 5 and go up from there with your potentiometer. Use a high impedance op-amp to buffer the input and output and provide a small filter to remove the f*32 switching glitches. A single RC hung on the output of the filters buffer amp works well. One look at the formula and you will see that doubling the filters frequency doubles the Q for the same resistor.
Be careful of too much Q. The response of the detector will become way too slow, Dave. * * *
 
Hi Dave
Thanks you just coured my headache, very impresive filter and in my more 20 years with electronics I have not even heard about it.
Thanks a lot
Mark
 
Eric,
You are dead right as to what the filter does. It is indeed possible to delete a number of the capacitors and use only the ones needed. I did not bring this up before as it would have caused confusion. Here is the heart of the matter, and how to use the filter. The coil amplifier uses two stages of amplification with a comb filter between the the first and second stages. The comb filter removes noise and interference from the output of the first amplifier. This stops the second amplifier from overloading in the presence of a strong noise source. By setting the gains of the two amplifiers correctly, you can now amplify the received signal to a much higher level than is possible with an untuned amplifier. The high gain of many PI amplifiers I have seen, caused them to output square waves when the coil got near a strong source of interference. The dual comb tuned amplifier allows much more amplification than a regular PI can ever hope to provide. It is also of course an order of magnitude easier to work on a PI design which provides a response which is free of noise. While I have to admit that one can use a differential input amplifier and a number of other methods to reduce the noise problem, the above method is very difficult to match.
 
Hi Dave
What about using different sizes af capacitors, I mean why not chose some bigger ones in the period you expect to be zero's ?.
Mark
 
Mark, My hat is off to you. Your idea is a good one. The sections of the filter corresponding to the part of the waveform that you expect to change rapidly could have small capacitors while the sections where only small changes are expected could have larger caps. You could also use more than two values of capacitors for a graded response. Congratulations, you just invented a sub cycle Q Vs time filter! How about this, A PI using a microprocessor could select different R values for the filter by addressing an added multiplexer which switches the resistor value. This would give a variable Q control to a plurality of filter positions. The micro could do this under program control based on what it is measuring from the signal *** Most of my detectors have been of the induction balance type. The comb filter is very nice as it does not mess up the received signal phases. I use two modified comb filters. One passes only the in phase components of the signal and the other passes only the quadrature signal components. The signals correspond to the R and X signals which are normaly seperated by synchronous demodulators. This is a great way to build a multi-frequency machine as the clock tunes the filter. You do however have to change the input resistor to compensate for the change in Q with frequency. A really powerful use for the comb filter is to use it as a notch filter. The input signal goes to both the filter and an inverting amplifier with a gain of -1. The output of the filter is summed with the amplifiers output. This provides a notch which unlike any other filter, will remove any repeating waveform. Very complex repeating waveforms can be notched out this way. Using this method it should be possible to filter a PI's response with a high Q comb filter and sum the filters output with the inverted direct signal leaving you with only the change in the PI's response as the coil moves over a target. I have not tried this and I dont know if it would be of any practicle use? I have included it only as an illustration as to the possibilities of this type of filter. However, it might be an idea for a PI motion detect mode? Let's keep the ideas coming. I will post a full schematic of the comb filter as soon as I get a chance, Dave. * * *
 
Hi Dave
I am glad you like my idea about using different sizes of capacitors. Yes, the idea using a microprocessor to tune the filter by R or C (or both) value should be possible and might be what we all are looking for (read:DISCRIMINATION). So thanks again for bringing a good brain exercise into this forum after a long summer vacation.
Mark
 
Hi Mark and Dave,
All sorts of possibilities come to mind. You only need closely spaced samples to resolve the front part of the waveform. As the waveform decays the sample width could increase accordingly, in fact, this is a technique used in Time Domain Electromagnetics for mineral prospecting. Taking your 32 position switch, the sampling could be done in a binary progression where each sample was twice the length of the previous one. Say the clocking was such that the first position on the switch had a duration of 10uS; this could be the delay after the end of the TX period. Position 2 would also be 10uS and would be the first sample. Position 3 would be strapped to position 4 to give a 20uS sampling period. Position 5 would be strapped to 6,7 and 8, while position 9 would be strapped to all positions up to 16. 17 would then go right through to 32. As the duration of each total sample is twice that of the previous, the time constant has to be adjusted so that all five are equal. This could be done by increasing the capacitor value in each stage. e.g. 0.1, 0.2, 0.4, 0.8, 1.6uF. For higher resolution, you could use a simple 1x, 2x, 3x, 4x....7x progression, leaving 3 unused positions at the end.
Eric.
 
You must log in or register to reply here.
Top