Some more questions....

[Anonymous]

1)I've heard it said that many PI operators can determine the ferrous composition of a target by the width of the signal, that is, a wide and slow rise time for ferrous and a narrow and steep rise time for non-ferrous.
Is there a direct correlation to the reflected pulse decay and the signal width?
2)Rather than determining a signal output from just one pulse and it's reflection, would it be possible to fire a train of pulses (10 maybe) and with each succeeding reflection, retire the receiver window. This would produce a set of co-ordinates defining the decay curve. Analysis of this curve through a differentational algorithm would allow approximation comparison of such with an on-board table. Each value in the table would have an assigned tone, thus the processor would synthesize an audible tone, it's pitch determined by the decay curve characteristics of the object. The higher the tone the more non-ferrous and visa-versa.
Is it worth pursuing?
I need a good design project for next year.
Thanks. Regards Kevin

 

[Anonymous]

[No message]

 

[Anonymous]

Eric has a tech forum at:
http://www.insidetheweb.com/mbs.cgi/mb122618

 

[Anonymous]

[No message]

 

[Anonymous]

[No message]

 

[Anonymous]

Hi Kiwi
I am doing something like that, I have a comperator that I can change the threshold point with a uP. I then have a counter that counts how many cycles the uP has to wait for the threshold point to be reach from the reflected puls. By measuring different counter values at different threshold points I can measure the rise (fall) time of the reflected puls. It also means that I can start measuring much earlier on the decaing puls, I do not have to wait until its close to zero. I am using a much faster PIC like uP called Scenix SX28AC it runs on 50MHz compared to a PIC16C84 thats run on 4 MHz, this means that I can do one instruction in 20 nS compared to 1uS.
My first eksperiment convince me that its the way to go. What you have do is fine the relation between the decay of the dying curve falltime and the time before threshold reach zero, and maybe some more factors I have not realised yet.
Mark

 

[Anonymous]

Hi Kevin and All,
It

 

[Anonymous]

Hi Kevin, Jeff and Mark,
There are many possible ways of analysing the return signal in a PI. You could certainly take a series of ten transmitter pulses and retard the delay on each pulse to progressively scan the decay curve and repeating the sequence ad infinitum. The more usual method is to have, say, ten sample pulses following each TX pulse and do a parallel processing of the data which is faster. It is also beneficial to have more samples at the early delays than at later times to determine the curve more accurately. This is done quite simply by having each succeeding sample pulse twice the length of the previous. i.e. if you have an initial delay of 15us followed by a 15us sample pulse, the next sample pulse would be 30us wide starting at the termination of the 15uS sample. The next sample would be 60uS, then 120 and so on for as many channels as you want before the next TX pulse. Personally, I think that nine or ten channels of binary spaced samples is sufficient to accurately determine the form of the decay curve. However, there is no harm in trying more and those with expertise in DSP may well have an argument for many more samples. About three years ago we did a project with a university computer science lab on digitising the decay curve after each pulse. They took over 100 samples starting before the back emf had reached zero and then doing digital filtering and comparison with look up tables. The results showed promise on a limited range of objects but the detection range was not very good due to signal to noise problems. DSP is an area where there could and should be a breakthrough, but it has not happened yet for PI detectors and it is noise that is the killer. In multi-sample systems each succeeding sample is diving toward the ambient noise level which can be horrendous in a broad band system. Certain noise filtering can be done at the front end but you have to be careful that the decay curve information is not distorted by the filter. The noise can be reduce by using longer integration times but the response of the detector then becomes slower. So, it is not an easy nut to crack but I

 

[Anonymous]

I get the impression that the decay curve analysis may be a method with a large margin of error, and requiring massive processing power thrown at it.
From what you say Eric, it may be more feasible to investigate DSP of the audio signal before the final output stage?
I had another idea, have you got any info as to whether or not this would work?
Utilize the magnetic properies of ferrous objects by using a series of pulses. First of all transmit one long pulse without sampling the reflection, this will begin to align a number of affected domains, then transmit a normal pulse and sample at a couple of regions, hold the value of this reflection. Now transmit another normal pulse and sample at the same regions, but this time reverse the current flow in the coil.
Now compare the value of this refection with the first, if the object is ferrous it may have a lower value as energy is expended in realignment of the magnetic domains. If the object is non-ferrous the two samples will be nearly identical.
This would require perhaps dual recievers, gating and maybe opto-couplers, along with noise reduction. Would there be enough of a difference, (if any) remaining from a ferrous object after noise reduction to allow discrimination?
Thanks and regards Kevin.