On Thu, 22 Jul 1999, Russell McMahon wrote:

> Summary: Using a very cheap op-map ($US0.16 in true volume) and a few
> relatively cheap parts and an (internal to uP) comparator an Analog
> to Digital conveter with 9+ bits accuracy is achieved. Better
> accuracy will be achieveable with longer conversion times. The
> technique is slow but cheap to implement.
>
>
> Must be time to stop playing with this.
> Results keep getting better.
> Many things that people say don't seem to be true!

The optimistic things or the pessimistic things? If optimistic, then I
have some lies to tell.

I've made some measurements of the 2-I/O line sigma delta A/D converter
and have been able to obtain fairly good results. I made measurements over
the range of about 0 to 1.4 volts and found that the maximum linearity
error was only 0.033% (assuming the data spans the full 4096 counts, which
it doesn't - I really need to measure over the full range to get an
accurate estimate of the linearity). This corresponds to about 1.3 counts
on a 12bit A/D converter. Here's the raw data I measured:

-0.0098   803
0.0484    831
0.0876    852
0.1133    867
0.1480    883
0.1827    89e
0.2000    8ac
0.2411    8cd
0.2871    8f2
0.3085    903
0.3450    920
0.3924    945
0.4385    96a
0.4708    984
0.5227    9ad
0.5620    9cd
0.5840    9de
0.6329    a05
0.6827    a2e
0.7086    a42
0.7654    a70
0.8133    a97
0.8783    aca
0.9158    ae8
0.9566    b08
0.9915    b24
1.0388    b4a
1.0822    b6d
1.1259    b8f
1.1756    bb8
1.2197    bdc
1.2454    bf1
1.3044    c1f
1.3332    c37
1.3895    c62

I also ran your (Russell's) data through the octave program and got
similar results.  Which is to say, that the linearity is quite good!


The latest thing I added was a simple low pass filter that averages the
last four samples. This routine is useful for other things too (such as
that median filter algorithm we beat too death a while back):

;----------------------------------------------------
; fir_average - find the average of the last n 16-bit
;               samples
;
; C:
;
; int m[SAMPLES]={0,0,0,0 ... };
; int
;    sample_number=0,
;    sum = 0;
;
; int fir_average(int new_sample)
; {
;   int w;
;
;   /* Find difference between the new sample and
;    * the one we are about to replace */
;
;   w = m[sample_number] - new_sample;
;
;   /* Subtract this difference from the old sample.
;    * This places the new sample into the array.
;    * e.g. w = x - y followed by x = x - w will replace
;    * x with y.
;    */
;
;   m[sample_number] -= w;
;
;   /* Subtract this difference from the total too. */
;   sum -= w;
;   if(++sample_number > SAMPLES)
;     sample_number = 0;
;
;   return(sum/SAMPLES);
; }
;
;inputs:
;  INDF points to &sample[sample_number]

fir_average:


        movf    x_lo,w
        subwf   INDF,W
        skpnc
         decf   sum_hi,f
        subwf   INDF,f
        subwf   sum_lo,f
        skpnc
         incf   sum_hi,f
        incf    FSR,f
        movf    x_hi,w
        subwf   INDF,w
        subwf   INDF,f
        subwf   sum_hi,f


        return



This routine is used like so:

        incf    sample_number,w
        andlw   0x03            ;only four samples
        movwf   sample_number
        addwf   sample_number,w
        addlw   s0l             ;Start of sample array

        movwf   FSR

        call    fir_average



>
> Latest trial uses 2 transistors to act as current sources to drive
> the integrating capacitor.
>
> Here's some shocking ASCII art (I hope)(I can see it - YMMV)
>
>           ______________________  Vcc
>                     |    |
>                     R    R
>                     R    R Re-hi
>                     |    |
>                     |    E
> PIC Output --.-RRR--.---B     Q1 - PNP
>              |           C
>              |           |              Rin
>              |           -------------RRRR--- <-- Vinput
>              |           |          |
>              | Rbi-lo    C          |
>              .-RRR--.---B   Q2     ---  Cint
>                     |    E  NPN    ---
>                     |    |          |
>                     R    R  Re-lo   |
>             Rbg-lo  R    R          |
>                     |    |          |
>  ------------------------------------- Gnd


If you're using the LM324 as a buffer amplifier (which I think is
necessary for almost every application [the amp not the 324 per se]) then
why not do something like so:

  |       /|
  |      /+|----------<< Vin
  |     /  |
  +----<   |
  |     \  |
P |      \-|--+
I |       \|  |
C |           |
  |           |
  +--/\/\/\---+
  |           |
  |          ===
  |           |
  |          ---
  |          ///

The 324 is being configured as a comparator. If Vin is greater than the DC
voltage on the capacitor, then the 324's output is high. The only problem
I see with this approach is that the voltage on the capacitor is a
function of the PIC's supply voltage - which may be too noisy or too
inaccurate for a 12-bit A/D converter. (This problem exists with the 1 and
2 I/O line a/d converters too, btw.) The programmable current source that
you show above addresses this issue if it's powered from a voltage
reference or perhaps a well filtered version of the power supply. It's
possible to also design a precise current source using a LM324.

Scott