Digital PI project first results

[Anonymous]

I don't have a PI board yet, but I have my first results from the digital PI project by connecting the PI processor to my VLF board. To keep track of which board is which, I will call the new processor board the PI DSP (Digital Signal Processor).
I connected the logical coil drive signal from the VLF board to an external interrupt line on the PI DSP. And I connected the output of the VLF amplifier to the analog input of the PI DSP. The VLF frequency is approximately 6600 Hz (158 usec period). You can think of this as a PRF of 6600.
On the falling edge of the coil drive signal an interrupt occurs. A few microseconds after the interrupt I read the A-D and add the sample into an integrator. I also start a timer that will go off 180 degrees later (79 usec). When the timer goes off I read the A-D again and subtract this sample from the integrator. This gives one differential sample per cycle for the integrator.
I count 220 cycles and then send the output of the integrator to a D-A converter. So I can see the output of the integrator on a scope. This gives samples at about 30 Hz. Later I will be filtering these samples.
The integrator is a low pass filter with a time constant of 256 cycles which is about 40 msec.
The output from the PI DSP looks about as good as the output from the VLF board which has a lot more filtering. The PI DSP has a couple of advantages over the other board. It has a 12 bit A-D vs. 10 bit. And because of the faster A-D it can get one differential sample per cycle while the VLF board only gets one differential sample every three cycles. So the PI DSP is getting more samples with more bits per sample.
This is not much of a test though because the VLF amp has a much narrower bandpass than a PI amp. I will be seeing a lot more noise at the A-D when I put this on a PI board.
Robert

 

[Anonymous]

Good for you Robert. By the way, the trick with delaying the pulse by one period and feeding it back to be added to the next pulse needs to be done ahead of the A/D converter. Thereby overcoming the problem of signals that are below the threshold noise of the A/D converter. Is that what you were planning on doing?
Randy Seden

 

[Anonymous]

Randy
If the noise at the input to the A-D is less than the LSB then you would have to do something before the A-D to see signals below the noise. Otherwise there would be certain ranges of input signals that would all give the same output from the A-D. Integrating many copies of the same number does not give any additional information.
If the noise at the A-D is always greater than the LSB then it is not necessary to do anything before the A-D. Any input signal level will always produce a noisy output from the A-D. The noisy digital signal from the A-D includes information about the input signal level. That signal information can be extracted by integrating or filtering the digital samples.
For example, suppose that you have a noisy signal level such that out of 100 A-D samples half will have a value of 5 and the other half will have a value of 6. When you add the 100 samples together you get a result of 550 which represents an input level of 5.5. Then increase the signal level by 1/10th of the LSB. In a noiseless system a change of .1 LSB might not cause any change in output at all. But with 1 LSB of noise the 100 A-D samples would give about 40 fives and 60 sixes. When you add these up you get 560 which represents an input level of 5.6.
The integrated signal still has noise in it so you will not get 560 for every 100 samples, the results will jump around. But by integrating N samples you decrease the noise by the square root of N. So if the input to the A-D is a signal of 5.6 plus noise of 1. the output of the integrator is 560 plus noise of sqrt(100)=10. If you divide by 100 you get a signal of 5.6 plus 0.1 noise. That is about 10 times better than you get by taking just one sample.
Robert

 

[Anonymous]

What I had in mind was an analog (read Continous Time) delay line taking the output of a Dual Gate Mosfet (off the source) and feeding it into one of gates, whilst the input signal feeds into the other gate. Thereby megohms of port to port isolation is achieved and the signals will algebraically add up and noise not in phase will decay as you mentioned.
The delay line would be a simple opamp variety as the R and C values are easily matched. I have also had incredible results using the Panasonic Bucket Brigade (CCD type) delay line. Made a super comb filter using these where the input is split into 2 paths, one thru the delay line and into one side of a diff. amp and the thru path directly into the other side of the diff amp. And guess what I used for a diff amp. in a pinch? An LM386 Audio Amp!!
If your not familiar with designing analog Time Delay Filters I'll be most happy to design one for you.
Randy Seden

 

[Anonymous]

Randy
Someone used to make short delay lines with a tap on every stage, about 12 or 16 stages. These were good for making sampled data filters. I tried to search for one a few weeks ago without any success. I don't know if they stopped making them, or if I just could not remember the right terms to search for. This was about 30 years ago and I don't remember what they were called. Do you know of any part numbers for this type of short CCD delay line?
Robert

 

[Anonymous]

Yes, I still have the Data Book at home. The company was EC&G Reticon and they USED to make the Transversal Filter chips. EC&G is long gone like most Aerospace related companies :>( .
So your best bet today would be the Panasonic BBD Chips that are sold by Digi-Key for around $6. Call and get the data book as it's very user friendly. The 2nd Gen. IC's are much quieter too and Digikey carries both versions so you've got to be carefull when ordering. Hope this helps.
Randy Seden
P.S. My forte is analog signal processing (and RF design)and I can see your a digital guy. So maybe you can help me cross the knowledge barrier to DSP as I'm stuck in the 1970's !

 

[Anonymous]

It's "EG&G" (not EC&G), and they still exist, though they no longer make anything.

 

[Anonymous]

In the 70's I was a digital guy. Now I am strictly a software guy.
I did do a few designs with 709 op amps in the old days, but it was never really my thing.
Robert

 

[Anonymous]

Joe,
Good work. It was the Reticon Div. that made the Analog CCD Delay Lines and turns out they were recently aquired by Perkin Elmer which makes ALOT of very cool electronics. Unfortunately CCD delay lines are not one of them!
Robert, I dug around in my lab and low and behold I still have a Reticon P/N SAD1024 CCD Delay line chip that you are most welcome to. I also have a Plessey MS1013 CCD Delay line chip that was made for video so it has a bandwidth of 5MHZ that is yours. The 1013 was a lower cost version of the 1003 which is a 910 bit CCD analog Delay line.
I DO have the Data sheet for the Plessey, but not for the Reticon chip. Maybe online somewhere. Radio Shack used to sell them believe it or not.
Let me know.
Randy Seden

 

[Anonymous]

Randy
Thanks for the offer of the delay line chip, but I was not really planning on using one myself. Once I started looking into PI designs I felt that the designers are not making full use of all the technology that is available.
I plan to take the DSP approach. I was thinking about delay lines for people who are more comfortable with op amps than with CPU's. Of course they could design sharp cutoff filters with op amps too, but it takes a lot of components. I was thinking about ways to improve filtering without adding too many components and without having to learn how to program a micro controller.
A delay line with taps, some resistors for coefficients, and an op amp to sum up the terms makes a simple filter that seem to be a natural for PI's which are inherently sampled systems. I just prefer to do it with software.
Robert