A/D data acting peculiarly.

[JimB]

I am attempting to generate values to plot a curve. I am reading a waveform that has the general shape of an exponential decay. I take a 32 reading average at each point then increase the delay 1 usec and take another 32 average. Instead of the data recreating the curve, it appears to take four identical readings then another four identical and so on. I have examined the electronics for synchronous noise and have filtered the reference very heavily which appears to be working since the reference is stable and quiet when viewed on the oscilloscope. The data for a couple of runs is shown below and the code follows. I don't think that the problem is in the hardware but something peculiar in how the time delay is incrementing (or not). The frequency of the source from the micro can be changed but the ultimate results are always similar.

Data read out:
Delay usec Value Value
43 355 358
44 355 345
45 343 373
46 307 309
47 308 309
48 307 309
49 307 309
50 279 281
51 279 281
52 279 281
53 279 281
54 260 265
55 263 258
56 263 265
57 257 265
58 246 247
59 246 247
60 246 247
61 246 247
62 234 235
63 235 236
64 234 236
65 235 236
66 227 226
67 224 228
68 227 225
69 224 226
70 217 218
71 217 218
72 217 218
73 218 218
74 211 212
75 211 212
76 211 212
77 211 212

[PCM programmer]

I looked at the delay_us() code in the .LST file, and it looks like CCS
has decided that at your clock frequency, the granularity of that
routine will be 4 us. One quick way to fix this is just to drop in
a 20 MHz crystal. Change the #use delay() statement and also
go to HS mode. Then you should get 1 us resolution in your delays.

--------------
Also there are a few anomalies in your code that don't affect this
problem but should be fixed.

1. You have an #int_ext statement but you don't have an isr.
The statement has no effect and should be removed.

2. You have two comments where you give the hex value of
a decimal number. The hex values are incorrect.

3. In the #use rs232() statement you probably should add the
ERRORS parameter if you're going to add code to receive data.

4. For a 5 MHz crystal, you should use the HS fuse, because the
16F877A data sheet says the max frequency for XT mode is
4 MHz. This spec is given in Table 17-3 in the Electrical
Characteristics section of the data sheet.

[Guest]

How interesting, regarding the 4USec resolution. What clock frequency would one get the 1USec resolution back? I am using the 5MHz to save power since this is a battery operated project. Is there some algorithim I could use to determine what the resolution would be as a function of clock frequency? I was thinking about lowering it even more to decrease the current demand.

Will remove the unnecessary code. Comments were OK before the code massaging started. They just haven't been kept up to date.

Another problem I am having is that whenever I use a fixed delay (no curve tracing to be done.) my data sill has a cyclical beat to it. If the time delay is fixed, no matter what delay or PWM frequency used, the results always have a sawtooth pattern. From what you have discovered, could that 4USec uncertainy be a part of this problem also? If so how would one get around this problem? In otherwords would data taken with a fixed delay be modulated by a 4USec dither or uncertainty that at the same time be regular enough to produce this very repeatable variation in readings?

[treitmey]

Is it possible to use dual freq? High speed for reading, then to low speed for sleepy time. or are you taking readings all the time?

[Guest]

Since I am using the processor at 3V, Microchip states that the frequency must be limited. There is a curve in the data sheet that shows this relationship.

But aside from this delay_us() problem, my cyclical data taken at fixed delays is quite troublesome. The output data, after plotting, has this sawtooth variation which is very stable and repeatable. It does change for various delays but there always seems to be this oscillatory pattern.

[PCM programmer]

[Guest]

The battery is actually a 3.6V cell, but is now running at about 3.16 volts and it does run.

[PCM programmer]

[JimB]

PCM Programmer,
I checked my programs this evening. I have two that I am working with and sure enough the BROWNOUT is listed in the FUSES directive. The battery voltage is at 3.156 V and the programs do run. Could this be a LF quirk?

Also, I queried CCS support about the 4USec granularity that you observed and also questioned them about my cyclic data readings and in addition to when the new ICD-U would be available. Here is their reply.
***********************************************
"1. The resolution if you use constants in the call to delay_us is 1 instruction time. With a 5mhz clock your instruction time is 0.8us. The delays are created by executing a certain number of instructions.

If you have a variable for the parameter to delay_us() then a call is made to a function that executes a loop for each us. In this case the fastest loop we can do on the PIC is 3 instructions. However in order to reduce the overhead the math is kept simple so the compiler may do something like divide the requested us by 4 and make a 4us loop. 4us is exactly 5 instructions making it more than the min 3 and allowing fast math (shift by 2) on the us variable.

2. What value to you use for the ADC clock (in call to setup_adc())? Make sure it is long enough for the proper conversion. Are you reading only one channel? If not make sure you have a delay after the channel switch long enough for the internal cap to charge. Does your analog source and the PIC operate off the same voltage regulator? If not you may be seeing the voltage difference between the two regulators if the error is within the sum of the rated errors for the regulators.

3. They are being built now. I think we will have them in a week. You should ask for the Rev 5 units."
**************************************************
Not quite sure I understand everything discussed in answer 2. Nothing applies so I am still puzzled by this very regular output.

[PCM programmer]

[JimB]

1. Vref is 2.440 Volts. This is supplied by the internal reference of a MAX5532, a dual D/A which has an internal reference voltage with a choice of 4 programmable values.

2. The reference voltage can supply up to 8mA.

3. The datasheet doesn't give the output impedance, just that it can supply the 8mA.

I changed the A/D clock as you suggested and that appeared to virtually solve the problem. I still get some variation but no more than 4 counts. I tried all the delay_us() values for which the curve exists and saw that the cyclical noise only showed up near the beginning and end of the curve. The middle section read with a solid reading.

The curve is generated by having a custom designed probe in a conductive solution and is the result of polarization of the electrodes.

I then replaced the probe and solution with a decade box and all readings were solid, i.e. no deviation, not even 1 count.

Since this appeared to work could you expand a little on what is going on? I have not really understood how the A/D clock is generated especially the ADC_clock_internal statement.

Thanks for you very helpful input.
JimB

[PCM programmer]