Simple Maths Going Wrong

[Darrel Taylor]

Hmm, I don't know Bill.
I've had a play with your code, and I'm not sure about it.

It does seem to pick a valid point in-between two samples. But it only averages that point, and the previous point found between two other samples. It's not averaging all the samples, so it's still very jumpy.

Your explanation describes EEPROM locations up to 29, but there are only 15 DATA values, although I don't think that affects the problem.

I've modified the program a bit to use the Average as the X value, then read the samples one at a time, also connected a POT to simulate the wind vane, but I get the same results.

Another hiccup, if the average is currently 10, then it receives a string of 255 samples, it stops at 1, and never crosses 0. And for many other samples, it stops 1 number away from the actual reading. 250 might stop at 249 or 251.

I think finding a point in-between 2 samples is fairly easy. But averaging a large number of samples is something altogether different, and a 2-sample average just doesn't do it.

Aratti,
The ATN function is actually an ATAN2 type since it works for all four quadrants. An arctangent would only work with 1 quadrant.

The format of the statement is ...

Result = X ATN Y

There aren't any other "math signs" involved.
<br>

[Bill Legge]

DT. Thanks for taking time to look into the code.

First, I wrote it specifically to average 16 samples - it won't work with any other number because of the counter:

J/2 + 16

This counter takes the average (or average+128 if the two sample are more that 128 apart - and checks for a 'roll-over' above 255) and stores them:

Samples 0 and 1 average stored in address 16
Samples 2 and 3 average stored in address 17
Samples 4 and 5 average stored in address 18
Samples 6 and 7 average stored in address 19
Samples 8 and 9 average stored in address 20
Samples 10 and 11 average stored in address 21
Samples 12 and 13 average stored in address 22
Samples 14 and 15 average stored in address 23

Then, the same J/2 + 16 deals with the next round of averages:

Averages of 16 and 17 stored in address 24
Averages of 18 and 19 stored in address 25
Averages of 20 and 21 stored in address 26
Avergaes of 22 and 23 stored in address 27

And again

Averages of 24 and 25 stored in address 28
Averages of 26 and 27 stored in address 29

And finally

Averages of 28 and 29 stored in address 30

I guess the same could be done for other numbers of samples but 16 seems a reasonable number for wind direction and the code is short and simple. I have spent a laborious few hours with a pocket calculator checking the results - and so far it is Ok. If you, or anyone else spots an error please let me know.

Regards Bill Legge

[Darrel Taylor]

WOW! I really got that interpretation wrong didn't I?

OK, I think I understand what you're doing.
I'll play with that a bit and see what happens.

Thanks for explaining.
<br>

[Darrel Taylor]

I'm thinking you might have a winner there Bill.
But I have a question or two.

I've had another go at your code and am getting some promising results.

It's hard to follow with a calculator, so I made an interface to help test the routines.

Also, instead of reading/writing to EEPROM, I've modified your routine to use an array, so that I can pass it data from the PC.

The array is a circular buffer that overwrites the oldest value with the new sample, and on each iteration runs your code to average all 16 samples in the array.

I'm thinking that the circular buffer might be a problem.
Maybe if it only took 16 samples then averaged ???

The modified routine looks like this ...

X  VAR BYTE                ' First vector
Y  VAR BYTE                ' Second vector
Difference  VAR BYTE     ' Difference between vectors, always +
Average  VAR WORD      ' Average of two vectors
J  VAR BYTE                ' EEPROM memory address

BillsData  VAR BYTE[32]
DataPos    VAR BYTE

Bill_Avg:
  DataPos = DataPos + 1
  IF DataPos = 16 THEN DataPos = 0
  
  BillsData(DataPos) = DIR

;DATA 250,230,240,34,200,10,10,23,23,21,125,0,34,255,123  ' 16 vectors	
  FOR J = 0 TO 28 STEP 2	' Read pairs of vectors
;   READ J,  X
    X = BillsData(J)
;   READ J+1,Y
    Y = BillsData(J+1)
  	if Y>X THEN SWAP X,Y	' X is always the larger variable
  				
  	Difference	= X-Y		
  	Average	= (Difference/2) + Y' Half difference and add to smaller vector
  		 	
  	IF Difference>128 then					
  	Average=Average+128	' Reverse direction of vector
  	IF Average>255 THEN Average=Average-255	' Whole revolution? Subtract 255 if necessary
  	ENDIF
  	
;    WRITE J/2 + 16,Average
        BillsData(J/2 + 16) = Average
  NEXT J
	Bill_Value = Average
;	LCDOUT $fe,$80,"Average: ",DEC3 Average	
RETURN

The interface simply sends the Wind Vane readings to the PIC where it runs both of our averaging codes, and sends the results back for display.
Yours is on the right. Any thoughts on the sample buffer?
It works great unless there are large fast changes. (click image to play)

<object id='stUkhdQ01IR11XRFVbWlleU19R' width='711' height='444' type='application/x-shockwave-flash' data='http://www.screentoaster.com/swf/STPlayer.swf' codebase='http://download.macromedia.com/pub/shockwave/cabs/flash/swflash.cab#version=9,0,115,0'><param name='movie' value='http://www.screentoaster.com/swf/STPlayer.swf'/><param name='allowFullScreen' value='true'/><param name='allowScriptAccess' value='always'/><param name='flashvars' value='video=stUkhdQ01IR11XRFVbWlleU19R'/></object>

Sorry Internet Explorer only.

[Bill Legge]

Darrel,

Thanks again for your investigation. The display 'InstruMental' is really good.

I guess the issue of:

1. Adding one new vector and then running the 'Averaging Tree' with 15 older vectors. Or...

2. Waiting till 16 new vectors are available and then running the 'Averaging Tree'

Is one of latency. And the selection should be based on:

1. The importance of immediate results.
2. The time interval between the availability of new vectors.
3. The need for the MCU to do other things.

I have attached some code that I used to send the averages to my PC with the screen presentation laid out so one can easily see, and check, the averages. The vectors come from a RANDOM function that loads up the PIC EEPROM. RS232 at 19,200 Baud on pin C.6. Loops continuosely.

Two questions for you please:

1. How does one post PBP code here so that it looks neat/tidy. My code looks OK on my PC but when I post it here it is a mess?

2. Is your 'InstruMental' writtern in Visual Basic? I want to learn this and recently got the book 'VISUAL BASIC FOR ELECTRONICS ENGINEERING APPLICATIONS' by Vincent Himpe. For me (An ex-electronics engineer) it's no good - starts off too advanced. Any advice on a good 'entry level' book?

Regards Bill Legge

' -----[ Title ]-------------------------------------------------
'
' File...... WVL 877 VERTOR AVERAGE
' Purpose... Experiment with PIC16F877A 
' Date...... July 2009
'
' -----[ Program Description ]---------------------------
'
' 1. Devices used:  PIC16F877A on EasyPIC5
'
' ----[ Includes / Defines ]-----------------------------------------
'
define OSC 20               ' Define Xtal as 20 MHz.  Use "HS" for programming
'
include "modedefs.bas"      ' Include serout defines
'
' -----[ Initialization ]--------------------------------------------
'
ADCON1	= %00001110			' A0 is analog, remainder digital
CMCON	= %00000111			' Comparators OFF
INTCON	= $80				' Turns off interrupts.  See PBP page 187
TRISA	= %00000001			' A0 is A/D input
TRISB=0 : TRISC=0 : TRISD=0 : TRISE=0                   
PORTA=0 : PORTB=0 : PORTC=0 : PORTD=0 : PORTE=0
'
' ----[ Variables ]--------------------------------------------------
'
X			VAR BYTE		' First vector READ
Y			VAR byte		' Second vector READ
Vector		var	byte[31]	' Stores 16 A/D readings of wind direction and averages
Difference	var	byte		' Difference between vectors, always +
Average		var	word		' Average of two vectors
J			var	byte		' EEPROM memory address
K			var	word
Memory		var	byte		' Contents of EEPROM address J
TxPin		var	PORTC.6
Heartbeat	var	PORTC.0
'
'-----[ Main ]-------------------------------------------------------	                           
'
Main:

	for J = 0 to 15									' Create 16 random numbers 1 to 255
		random K
		write J, K.lowbyte							' Load EEPROM with random numbers
	next J		
	
	for J = 0 to 28 step 2							' Read pairs of vectors
		read J,  X
		read J+1,Y
		if Y>X then swap X,Y						' X is always the larger variable
				
		Difference	= X-Y		
		Average	= (Difference/2) + Y				' Half difference and add to smaller vector
		 	
		IF DIFFERENCE>128 then					
		Average=Average+128							' Reverse direction of vector
		if Average>255 then Average=Average-255		' Whole revolution? Subtract 255 if necessary
		endif
	
	    write J/2 + 16,Average		
	next J
	
	for J = 0 to 30									' Send to PC at 19,200 Baud
		read J, Vector[J]
	next J
	serout2 TxPin,32,["VECTOR ADDITION",13,13] 
			
	serout2 TxPin,32,[dec3 Vector[0],9,dec3 Vector[16],9,dec3 Vector[24],9,dec3 Vector[28],9,dec3 Vector[30],13]
	serout2 TxPin,32,[dec3 Vector[1],13]
	serout2 TxPin,32,[dec3 Vector[2],9,dec3 Vector[17],13]
	serout2 TxPin,32,[dec3 Vector[3],13]
	serout2 TxPin,32,[dec3 Vector[4],9,dec3 Vector[18],9,dec3 Vector[25],13]
	serout2 TxPin,32,[dec3 Vector[5],13]
	serout2 TxPin,32,[dec3 Vector[6],9,dec3 Vector[19],13]
	serout2 TxPin,32,[dec3 Vector[7],13]
	serout2 TxPin,32,[dec3 Vector[8],9,dec3 Vector[20],9,dec3 Vector[26],9,dec3 Vector[29],13]
	serout2 TxPin,32,[dec3 Vector[9],13]
	serout2 TxPin,32,[dec3 Vector[10],9,dec3 Vector[21],13]
	serout2 TxPin,32,[dec3 Vector[11],13]
	serout2 TxPin,32,[dec3 Vector[12],9,dec3 Vector[22],9,dec3 Vector[27],13]
	serout2 TxPin,32,[dec3 Vector[13],13]
	serout2 TxPin,32,[dec3 Vector[14],9,dec3 Vector[23],13]
	serout2 TxPin,32,[dec3 Vector[15],13,13,13]

	goto Main	
		
END

[rsocor01]

Very nice animated simulation by DT. What program are you using to create this?

Now, in DT's posted program taking an average with "S=(S*(AvgCount-1)+Temp)/AvgCount" takes some time to converge. With sudden changes in wind direction, which is usually the case, a high sampling rate is needed to keep up with this algorithm. This is a common issue when averaging several readings.

Robert

[Darrel Taylor]

Bill,

After more testing, I can say that the problem I was seeing was due to my averaging the circular buffer after each sample.

By taking 16 samples (at high speed) then averaging, the vane can never go more than 128 unless it's crossing zero and everything works fine. If the samples are slow enough that the vane can go more than 128, the problem shows up again.

I'm sticking with SIN/COS/ATN.<hr>

The program I'm using is DT_InstruMental. Something I've been working on for awhile that makes it easy to display info from the PIC graphically.

It's still in it's infancy, but the only way to figure out what I need to do with it, is to use it. ... So I'm using it.

Here's one of my favorites so far ...

<object id='stUkhdQ01IR11XRl9VWV1aVF5T' width='345' height='308' type='application/x-shockwave-flash' data='http://www.screentoaster.com/swf/STPlayer.swf' codebase='http://download.macromedia.com/pub/shockwave/cabs/flash/swflash.cab#version=9,0,115,0'><param name='movie' value='http://www.screentoaster.com/swf/STPlayer.swf'/><param name='allowFullScreen' value='true'/><param name='allowScriptAccess' value='always'/><param name='flashvars' value='video=stUkhdQ01IR11XRl9VWV1aVF5T'/></object>