Digital PI results

[Anonymous]

The test conditions are:
Pulse repetition frequency: 5000 Hz
Main delay: 20 usec
Analog Secondary delay: 100 usec after start of main sample
Digital Secondary delay: 96 usec after start of main sample
Analog integration window: 20 usec
Digital sample: two A-D samples at each point, 8 usec apart
Integrating 4 times as many samples as usual
I am near the limits of what this microprocessor chip can do. I can probably get the number of samples higher if I work at it, but not a lot higher. To get some idea of what the results would be if I could get a lot more samples, I changed how many A-D samples the integrator adds together to get each filter sample. I increased it by a factor of 4. Since the number of A-D samples I am taking per second has not changes this means the filter sample rate has dropped from 30 to 7.5. This rate is too slow for a reasonable search sweep rate, but it might be useable for pinpointing where you are not moving the coil very fast.
All the gains from the previous post are the same except the integrator gain. The integrator gain has increased by a factor of 4 to 1280. If you multiply all the gains together you get the total system gain from amplifier input to integrator output (450 * .31 * 1.64 * 1280 * 1.22 = 357,000).
The system gain has increased by a factor of 4 so based on yesterday's sensitivity numbers a penny at 8 inches from the coil should give about 800 mV at the output. What I actually measure today is closer to 700 mV. There is a bit of slop in these numbers. At 9 inches the output is about 400 mV, and at 10 inches it is a bit over 200 mV. It is hard to make accurate measurements because with the system gain higher the noise has also gone up. But the noise has not gone up as much as the target signal. Today the noise is about 200 to 400 mV. So I will say the 400 mV target signal at 9 inches is the smallest that can be detected under these test conditions.
Dividing 400 mV by the system gain of 357,000 gives 1.1 uV at the input to the amplifier. This is close to the sensitivity of the analog PI board, but the analog PI gets that performance at a normal sweep speed while this digital PI can only do that at a very slow sweep speed.
Robert

 

[Anonymous]

These are some statistics about the processor and software I have been using for these tests.
The ADuC812 microconverter chip has an 8051 CPU core running off an 11 MHz clock. This chip uses 12 clock cycles for each instruction cycle. That gives about 900,000 instruction cycles per second. This is a bit slow by today's standards.
Most of the tests have been done at a pulse rate of 5000 pulses per second, and with two target A-D samples and two background A-D samples taken from each pulse. This is a total of 20,000 A-D samples per second. The interrupt routines that collect and integrate these samples use about 425,000 cycles per second or about 47% of the available cycles. The data is then filtered in a low speed loop which consumes another 10% of the CPU cycles. So all together about 57% of the CPU time is used.
I have left some time free because in a final design I would like to take another set of target samples for ID purposes. This would add another 15% to the interrupt routines and another 10% for filtering a second channel plus a little more time to compute an ID. That brings the total up to 82% which does not leave much room to spare. So I think that 20,000 samples per second is about all that this chip is capable of in a single chip configuration.
The filters are implemented as two pole sections that can be cascaded for more complex filters. Each section can be low pass, high pass, or band pass. Each section takes about 500 cycles to execute.
At 5000 pps and 2 samples and with approximately 30 mV of noise at the output of the amplifier I am able to detect a target that gives a signal of 1 mV at the output of the amplifier. That is, I can detect a signal that is 30 times smaller than the noise. The gain from the output of the amplifier to the output of the digital PI DAC is 200. So the 1 mV signal shows up as a 200 mV signal at the output. If there were no filtering at all, the noise at the output would be 6000 mV. What I actually get is 100 to 200 mV of noise at the output. The reduction in noise is due to the integration and filtering.
Robert

 

[Anonymous]

I know that I have not been testing very long and have not taken time to optimize the design, but my intention for phase 1 of this digital PI project was to determine whether I could take A-D samples of the amplifier output, digitally process them, and get results similar to the output from the integrator in an analog PI. I still plan to do additional testing, but I think I already have enough information to make that decision.
The first thing I need to say about these tests is that you should assume about a +/- 50% tolerance on any signal levels I have reported. And that is for the conditions under which I have been testing. If the same tests were conducted somewhere else the results would probably be different. I have seen day to day variations when I repeat the same tests, and on some days it hardly works at all. However, every time the results have gotten really bad I have been able to find a reason and correct it.
Working with 20,000 A-D samples per second I am getting about half the signal to noise ratio from the digital processor as I get from the analog integrator. That translates to about one inch less range for the digital PI compared to the analog. On some days I have thought it looked better than that, but averaged over the last week I would say the analog is better by a factor of two. A clean design tailored for this chip might do better than this, but I don't want to count on getting a 100% improvement just from cleaning up the design.
I think the ADuC812 is useable as a digital PI processor. I imagine that it would be easy to find 2:1 differences between PI boards from different detectors. But I would like to have some safety margin. To make the digital be as good as the analog I think I would have to use about 80,000 samples per second taken at half the interval I have been using. I have been able to take multiple samples about 7 to 8 usec apart. I think they should be taken 4 usec apart or less to get a signal equivalent to the continuously integrated analog signal. The number 80,000 is a rough estimate, I think something in the rage of 50,000 to 100,000 should be ok. To take samples 4 usec apart the ADC has to be capable of at least a 250 kHz conversion rate. The ADuC812 cannot go this fast but other chips can.
Robert

 

[Kev]

n/t