Need help to find out timebase ADC->UART ATMega64

Log in or register to post comments
Go To Last Post
4 posts / 0 new
Author
Message
lostchild
lostchild's picture
Level: New Member
Joined: Fri. Oct 10, 2008
Posts: 15 View posts
Location: Bonn, Germany
#1
Posted by lostchild: Tue. Apr 21, 2009 - 10:25 AM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Hi there!

I hope the subject is clear enough? Well, here is my problem. I use an ATMega 64 to do three major jobs on a PCB:

  • generate a clock for a switched capacitor filter (LTC1067) with the Timer1
  • convert the analog output behind the RMS-To-DC converter behind the filter through ADC0 to digital values
  • convert the values to ASCII and send them to an UART to a PC with a terminal program running and logging the results at 57,6kbit/s
The ATMega64 runs at 16MHz external clock. The program works fine, it filters ripple control signals from the main line and sends the ASCII values to an UART. BUT I don't know, how my time base is scaled. Here are some thoughts but I think they are not 100% right because the result I calculate with the pocket calculator is too far away from what it should be.
  • ADC runs at 125 kHz, so one cycle is 8us long. One conversion in free running mode takes 13 cycles, so 13 x 8us equals to 104us. In theory I should receive one result in my ring buffer every 104 us.
  • UART transmits at the speed of 57600 baud, that means 57600 bits per second which is 7200 bytes per second. My results in ASCII are 2-3 bytes long (one or two digits plus a CR) so in the worst case there are 2400 transmissions of ASCII value pairs every second which takes almost 417us per transmission.
  • that means I lose around every 4th ADC value because the UART is too slow to read the RingBuffer before it gets overwritten?
Here is the code to this program: Free running ADC

I tried to solve the problem in getting a defined amount of ADC samples. I thought I'd set up another Timer (Timer0) and trigger the ADC through Timer0. I tried to trigger every 200us, every 400us and even every 500 us.

  • every 200us seemed to look o.k. but the time base I calculated later did not match either. I assumed I trigger the ADC every 200us, on conversion takes 13,5 cylces in auto trigger mode. At 125kHz ADC clock this would be 108us. So I assumed I'd have one value every 200us. But this was not correct, I had less values. Maybe the INT to ASCII conversion and the UART takes too long?
  • every 400 or even 500 us I received "0" in between my values in my RingBuffer so I think the ADC was going on too slow?
Here is the code for the Timer0 triggered ADC attempt: Timer triggered ADC

I hope that someone can clear my mind up and help me how to get the correct timebase. I appreciate any help.

Josef

Attachment(s): 

File The first code

Tags:

AVR Microcontrollers, megaAVR and tinyAVR
Top
sternst
sternst's picture
Level: Raving Lunatic
Joined: Mon. Jul 23, 2001
Posts: 3499 View posts
Location: Rheine, Germany
Posted by sternst: Tue. Apr 21, 2009 - 11:53 AM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Ok, a few notes:

lostchild wrote:
  • ADC runs at 125 kHz, so one cycle is 8us long. One conversion in free running mode takes 13 cycles, so 13 x 8us equals to 104us. In theory I should receive one result in my ring buffer every 104 us.
  • UART transmits at the speed of 57600 baud, that means 57600 bits per second which is 7200 bytes per second. My results in ASCII are 2-3 bytes long (one or two digits plus a CR) so in the worst case there are 2400 transmissions of ASCII value pairs every second which takes almost 417us per transmission.
  • that means I lose around every 4th ADC value because the UART is too slow to read the RingBuffer before it gets overwritten?

57600 baud = 5760 bytes/s (you forgot start and stop bit).

If ADC conversation takes 104µs and transmitting the value takes 417µs, then you don't lose about every 4th ADC, but you lose 3 of 4 conversations (only every 4th result is transmitted).

In your code you use the UDRE interrupt and then you send multiple characters inside this interrupt with polling. Not a good idea in my opinion.

The ring buffer for the conversation results doesn't make much sense with your current concept. If you have a producer with a fixed rate at one end and a consumer with a fixed rate at the other end, the buffer will always be empty or always be full, depending on which end has the higher rate.

Stefan Ernst

Top
Gold Markclawson
clawson's picture
Level: Moderator
Joined: Mon. Jul 18, 2005
Posts: 104962 View posts
Location: (using avr-gcc in) Finchingfield, Essex, England
Posted by clawson: Tue. Apr 21, 2009 - 12:03 PM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Just to note that if your ADC produces faster than your PC (limited by baud rate can consume) you might want to consider averaging every 4 readings (or whatever the difference in rates are).

To speed up the PC link you might also want to consider not converting to ASCII. Say you are transmitting 10bits as 2 bytes, if you convert this to ASCII if could be over 999 so using 4 bytes. You might therefore double the rate (of course some ASCII readings might only be 1 or 2 characters too)

Top
lostchild
lostchild's picture
Level: New Member
Joined: Fri. Oct 10, 2008
Posts: 15 View posts
Location: Bonn, Germany
Posted by lostchild: Thu. Apr 23, 2009 - 08:25 AM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Thank you for your replies!

I worked on the program and re-wrote it a bit. I deleted the ring buffer part and followed clawsons advice to collect a few samples and put the sum to the UART. I did not average them on the Atmel, that job can do the PC much faster... All I need is a known time base which should be as fast as possible of course. The signal I would like to measure is approx. 150ms long. I did not really know how to send the ADC result to the UART. I tried to send it like:

  uart_puts(sum & 0x0ff); //low byte
  uart_puts(sum >>8);     //high byte

where sum is an uint16_t value. The terminal program running on the PC did not show numbers of course but weird characters so it is not ASCII what the Atmel transmits - sure. I'm not that experienced in programming - is it possible to read the values somehow on the PC through the UART even if the characters are NOT ASCII and convert them to ASCII on the PC?

I now tried to solve it in converting the number to binary digits and add 48 to them (convert them to ASCII) and send them to the UART. But I still do not know if the timebase is now 104us*5 (13 ADC cycles multiplied by number of values summed up). How can I make sure how long my program needs to run? I suppose not at all due to the interrupts I use and the compiler in between?!

Thank you very much for your support. I am looking forward to your replies.

Josef

EDIT: I put in toggle pins into my code and watched them with an oscilloscope. This way I found out how long my routines run. I got rid of any unnecessary stuff (like the blinking LED...) and summed up 4 values before I transmit them to the UART (now with 115200 baud) and it works now. I got a fixed timebase of 4x104us. Thank you very much for your help!

Josef

Attachment(s): 

File new version of the code
Top