[time-nuts] Tbolt temperature sensor

[Bruce Griffiths]

Mark Sims wrote:
> Actually,  what is apparently going on with the newer Thunderbolts is worse than that.  The reported temperature seems to only end in 0.25C or 0.75C,  so it has an effective 0.5C resolution.   The basic DS1620 resolution is 9 bits,  but the Tbolt firmware reduces that to 8 bits since the first step in the high res temp algorithm is to mask off the lower bit.  Whatever is going on  they are not getting the extra resolution that they think they are,  and in fact they are reducing the basic resolution of the chip.
>
> The firmware does seem do do some filtering on those values since whenever the reading steps you can see some smoothing  going on.  A lot of times the  temperature value oscillatates around the step point.  The filter apparently does not have any hysteresis.
>
> The older Thunderbolts produced a nice smooth curve.  The high res temperature reading (Bruce says is 12 bit/0.0625C) coupled with the firmware filtering gave temperature curves with microdegree scale resolution.  The newer ones clunk around with effectively 0.5C resolution.
>   
Not quite right.
According to the datasheet the resolution with the standard 9 bit output
is actually 0.5C.
The MSBit is actually a sign bit.
Despite what the text may perhaps implies, 1/2 of a LSB is actually
subtracted before combining with the residual count and other data (see
attachment for formula).
The resolution isnt in fact being truncated at all.
An offset of 1/2 LSB is being subtracted from the coarse reading, this
has no effect on the resolution.

Bruce
-------------- next part --------------
A non-text attachment was scrubbed...
Name: TQ9133.png
Type: image/png
Size: 50138 bytes
Desc: not available
Url : http://www.febo.com/pipermail/time-nuts/attachments/20090205/573b7007/attachment-0001.png