My PI project

[Anonymous]

I have decided to begin a digital PI project. There is a simple block diagram below that shows what I have in mind. I will sample the received signal with an analog to digital converter, then integrate and filter the samples to produce an output signal that can be modulated for a speaker.
I have completed phase 0 of the project. That is the part that does not have much to do with PI. I looked at some issues about sampling and digital filtering before deciding if I even wanted to try doing this. My conclusion so far is that if the amplifier bandwidth is about 10 to 20 kHz then I can directly sample the output of the amplifier, but I will be at a disadvantage compared to conventional PI's that have a sample window of 10 usec or longer. That is, I expect to have a worse signal to noise ratio on the raw samples than an analog PI would have. On the other hand I think I can do a better job of integrating the samples than do the few PI integrators I have seen. Maybe I can make up for the worse SNR with better filtering.
Phase 1 of the project will be to piggyback a digital processor on a normal PI board. I will pick off the coil drive signal to find the turn off time, then sample the received signal at certain intervals after turn off. The digital samples will be integrated and filtered and output for comparison with the output of the PI integrator. If I cannot get results at this point that are approximately equal to the output from the PI integrator then the project will be over. If I can get comparable results then I will go on to the next phase.
The 10 bit A-D in the processor I have been using in my VLF project is a little slow for what I want to do now. It is rated at 65 usec per conversion. I would like to be able to take samples closer together than that, and I don't want to load the board up with sample and holds. So I will switch to another board that has a faster 5 usec 12 bit A-D. Unfortunately, this board uses a different instruction set, so I will have to rewrite all my filter routines. Also I don't have a PI board to piggyback on yet. So I will have to see about getting one by the time I get all the software written.
Phase 2 (if I get that far) will be the diagram below, a single board in which the processor does all the timing, sampling, filtering, and as much of the audio as possible.
Robert

 

[Anonymous]

Robert,
When you say intergrate, is that where you delay a pulse by one period and feed it back into a summing amp in one port whilst the input signal goes into the other port and the two are added together?
This method (I was told a an old Hughes Aircraft Radar Engineer)is used in Radar signal processing and is called a"Video Integrator".
Definately would be a neat way to bring a low level signal out of the noise by simply adding it to itself.
Randy Seden

 

[Anonymous]

Robert,
I do a fair amount with DSPs and fast ADC/DACs (I'm doing 500,000 samp/sec at 12 bit for another project). Note that ADCs for this sort of speed (sub-1uS) are not expensive any more (< USD 5) - I use one from Analogue Devices that is available in a SOT23 package (AD7476ART). Note that you must be very careful about electrical noise during sampling even for 10bit ADCs - clean power (seperate from the digital supplies) and good ground planes & decoupling are essential. The real issue is what you do with the data and how fast you can process it.
The link below discusses a bit of one of my projects. I've only recently become interested in PIs and metal detectors per se, and have followed your tests with interest. I have designed a PIC-based PI detector with input from several existing designs - I went off the EE PIC design for various reasons, not least that it uses a PIC with loads of capability, and then uses it only as a glorified fixed pulse generator. I know there are other firmware sources, but its still very crude.
For testing stuff, I use an AD 2191M EZKIT with an Apex ICE - this works quite nicely for me, but its far from cheap to build a PI detector with this sort of technology - the PCBs are 4-layer, 8mil, for a start. The analogue side is not a problem, and as DSP projects go you are not talking high speed, so maybe something like a dsPIC would suit your requirements?
Regarding noise elimination/cancelling, you may find that using averaging and a good FIR will remove a lot of the noise quite simply. If you are being really keen, you could use short form FFTs to remove only the frequencies of interest - there are many possibilities...
Nicko

 

[Anonymous]

Nicko
I looked at PIC's when I was starting my VLF project and wrote a couple of routines for them. I decided I did not like it for signal processing. It is ok for simple timing functions but is too awkward for number crunching. I chose an AVR chip instead. Now I am working with a chip that has an 8052 instruction set.
Robert

 

[Anonymous]

Randy
Integration can be implemented many ways, but yes, you are describing a way of integrating sampled analog data. If the samples are digital you use an adder instead of a summing amp. Another way of integrating analog samples is to just dump them all into a capacitor and let it sum up the charges. This is how I have seen it done in the few PI circuits I have looked at.
The common thread is adding up a lot of samples taken at different times.
It does work well at pulling signals out of the noise. In my VLF project I can see 4 bits below the LSB of the A-D.
Robert

 

[Anonymous]

Check out the new Rabbit 3010 uPocessor core in a 1.2 X 1.2 package. CMOS runs on 3.3V.
It has 4 channels of 12 bit or 8 channels of 11 bit A/D (software controlled). It also has a clock rate up to 54 Mhz if you want to crunch some numbers. It is essentially a Z180 processor.