I'm using an AVR microcontroller to take readings from a MAX6675 (SPI K-type thermocouple ADC chip).

I am having difficulty understanding what is required by this chip's SPI interface. The datasheet (https://datasheets.maximintegrated.com/en/ds/MAX6675.pdf) is a bit confusing on the matter. Pages 5 and 6 discuss the signalling.

I'd like to take readings at a rate of about 16 times per second, one sample each from an array of 16 thermocouples (muxed together using analog mux chips). But the problem I'm having is that even though the analog muxing does work, it seems like the 6675 gets "stuck" sometimes and either repeats a reading over and over again or perhaps delays it by one channel's worth.

Here's the code I'm using to read the SPI data...

unsigned short ReadThermocouple(char channel)
    char i;
    unsigned short retval = 0;  
    //select the thermocouple channel...
    channel &= THERMOCOUPLE_ADDR_PINS; //keep the low 4 bits of the channel
    PORTC &= ~THERMOCOUPLE_ADDR_PINS; //clear the channel address pins (e.g. select channel 0 momentarily)
    PORTC |= channel; //select the actual channel requested
    Serial_Send_Char('A' + channel);

    //CS LOW
    PORTD &= ~SPI_SCn;

    for(i = 0 ; i < 16 ; i++)
        //raise clock line
        PORTD |= SPI_SCK;
        //lower clock line
        PORTD &= ~SPI_SCK;  
        //read the bit
        retval <<= 1;       
        retval |= (PIND & SPI_MISO) >> 2;   

    //CS HIGH
    PORTD |= SPI_SCn;
    return retval;

The problem looks to be that the data that gets returned seems to "lag behind" by one channel. E.g. I try to read channel 1 and I get channel 0's reading. Similarly after the first time through the 16-channel loop, channel 0's reading actually contains the temperature from the previous iteration's channel 15.

Obviously I'm driving the SPI incorrectly. But I can't figure out what's wrong.

The datasheet doesn't suggest how long it takes to record a reading. Perhaps I'm going too fast? It also does not suggest in detail how to properly arrange and synchronise raising and lowering the Chip Select line. The text of the datasheet and the timing diagram seem to contradict each-other.

I've tried all combinations of raising and lowering the various signals with different degrees of delays between them, but I can't get coherent output from this MAX6675, unless I go VERY slowly indeed, and then I still have this "off-by-one-channel" problem.

I've checked the physical hardware and it all seems fine, I'm driving the mux chips correctly. And I can prove that warming any given thermocouple with my hand does have a reliable effect on the readings. I'm fairly certain this is a timing issue or some mess-up with the sequence of events.

Am I driving this chip incorrectly? Am I asking too much of it to take readings at 16 times per second?

  • \$\begingroup\$ Where's the link to the datasheet? What are the relevant section(s) that are confusing you? \$\endgroup\$
    – Null
    Commented Oct 11, 2021 at 19:00

1 Answer 1


The ADC on the chip may make fewer than 5 conversions per second, as the datasheet indicates (0.22 second maximum time).

You can force initiation of a new conversion via the /CS pin as described in the datasheet. You would want the analog signal to be stable before you do that unless you intend on discarding the first conversion.

And yes, 16 conversions per second is definitely impossible.

Reading data over the SPI bus is a different (and potentially much faster) process than the measurement of a new data value. Your problem is not the SPI protocol per se, but rather understanding how the system works.

Suggested sequence:

  1. Change the MUX, wait for the signal to be stable (dependent on your circuit- if there is an RC filter it could be many ms)
  2. Initiate a conversion, wait for at least 220ms (or a bit more for safety)
  3. Read the data via SPI, and repeat
  • 1
    \$\begingroup\$ I see. I think I must have been assuming that the timing diagram for the SPI was asynchronous with sampling somehow. I suspect that I need to introduce a 300ms (or so) delay just after starting a conversion, and before I stop the conversion and begin clocking out the bits. Thank you. \$\endgroup\$ Commented Oct 11, 2021 at 19:34
  • \$\begingroup\$ @user882326721 Yes, and if I wasn't clear about the multiplexer timing, please ask. \$\endgroup\$ Commented Oct 11, 2021 at 20:48
  • 1
    \$\begingroup\$ Thank you, it is working well now. I was being greedy with the hope of 16 samples per second, my purpose needs only 1 sample per second technically. I've got my microcontroller feeding out an array of 16 temperature readings every four seconds in a human readable format (kind-of, the values are in hex, but it's good enough for long-term, efficient logging). So far I've only been using 4 thermocouples, now I've got to spend another £60 on 12 more to fill up the array! Why are they so expensive? Cheers :) \$\endgroup\$ Commented Oct 15, 2021 at 20:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.