simple LSB explanation please

If you treat the 12-bit value as an integer (0-4095), it can be up to +/- 8 from the correct value.

johnny78 said:

hi Guys
would you please explain what kind of number they mean by this(exhibits thermocouple accuracy of 8 LSBs)

regards
Johnny


The MAX6675 performs cold-junction compensation and digitizes the signal from a type-K thermocouple. The data is output in a 12-bit resolution, SPI-compatible, read-only format.

This converter resolves temperatures to 0.25°C, allows readings as high as +1024°C, and exhibits thermocouple accuracy of 8 LSBs for temperatures ranging from 0°C to +700°C.

The MAX6675 is available in a small, 8-pin SO package.

Click to expand...

what does it mean

std_match said:

If you treat the 12-bit value as an integer (0-4095), it can be up to +/- 8 from the correct value.

Click to expand...

if i want to display it as float how many digits after the decimal point i can get ?

johnny78 said:

what does it mean

if i want to display it as float how many digits after the decimal point i can get ?

Click to expand...

i was reading that the accuracy of LM35 is 1C & the TMP36 0.1C
so what is the accuracy of this converter ?

thanks in advance

johnny78 said:

i was reading that the accuracy of LM35 is 1C & the TMP36 0.1C
so what is the accuracy of this converter ?

Click to expand...

sorry its the same accuracy

& i think i can have a number for every 1mV so 2 digits after the decimal point
any better resolution idea please

thanks
Johnny

The resolution of the MAX6675 is 0.25°C, so 8 LSB accuracy means +/- 2°C.

Hello!

i think i can have a number for every 1mV so 2 digits after the decimal point
any better resolution idea please

Click to expand...

You cannot have a better resolution for free.
The thermocouple converter has a fixed range, apparently 0 ~ 1024 degrees. If you divide this
by 2^12 (or 4096), you end up to 1/4 degree, the 0.25 degree of the specs. And that's all.
You cannot have "a number for every 1mv". All what you will have is your reading +/- 2 degrees
in the 0~700 degrees range, and a bit worse above.

Now supposing that you have some random noise on your readings, you can use dithering in order
to get better resolution, at the expense of speed or delay. In this case, you can add 1 bit everytime
you multiply your samples by 4.

Dora.

tj

doraemon said:

Hello!



You cannot have a better resolution for free.
The thermocouple converter has a fixed range, apparently 0 ~ 1024 degrees. If you divide this
by 2^12 (or 4096), you end up to 1/4 degree, the 0.25 degree of the specs. And that's all.
You cannot have "a number for every 1mv". All what you will have is your reading +/- 2 degrees
in the 0~700 degrees range, and a bit worse above.

Now supposing that you have some random noise on your readings, you can use dithering in order
to get better resolution, at the expense of speed or delay. In this case, you can add 1 bit everytime
you multiply your samples by 4.

Dora

Click to expand...

thanks Dora

this is what i was thinking of
(In this case, you can add 1 bit everytime you multiply your samples by 4.)

do u suggest any better solution than LM35?
the lm35 can give me 10mv for every degree so i can have 2 digits resolution after the decimal point

thanks
Johnny

Hello!
Sorry, what I said is ambiguous:

In this case, you can add 1 bit everytime you multiply your samples by 4.

Click to expand...

What I meant is not multiplying samples by 4 which wouldn't make much sense, but increasing the number of samples by a factor of 4 and calculate the average.
In your case, you have 12 bit samples (0 ~ 4095). So if you take 4 consecutive measurements, it will add to a value in 4 times this range (0~16383). But as you increase the noise, you will go from 0~4095 +/- 8lsb to 0~16283 +/- 32lsb. Seen like this it sounds like you gain nothing since you have also increased the noise. But the 32 lsb deviation becomes extremely improbable. Most of the time, the errors will somewhat compensate. So you don't get 2 extra bits as you could first think by the amplitude increase, you don't get 0 extra bits as you would then think when you consider the +/- 32 lsb deviation. You get something intermediate, and the theoretical calculation bring you to +1bit per 4x sampling increase.

Dora.