Another question related to temperature sensors: why is the low pass filter frequency in most remote temperature sensors set to 65 kHz?

For example: ADT7485A, NCT65, ADT7467 etc

ywguo wrote on Jun 14^th, 2013, 8:47am:


Hi Yawei, thanks for your response. For a while I thought I wasn't going to get any

The reason I worry about glitches is if I decide to use an SAR ADC. Glitches at the output could lead to erroneous results, although it can be argued that if enough time is given for the output to settle, it should be ok.

More worrisome is the absence of an anti-aliasing filter. The wideband thermal noise will alias back in sampled data systems. Given that ∆Vbe increases by only 200-300uV per degree Kelvin, this noise could be an issue. That is why I asked about the need for anti-aliasing filters.

The non-ideality is introduced when an offset voltage is subtracted from the amplified ∆Vbe. Since the offset voltage comes from the bandgap, the amplified output will contain the bandgap error.

Finally, I wanted some comments on the use of the ADC. Usually a ΣΔ ADC is used in temperature sensors. Tuthill's paper mentions an SAR ADC. What is the advantage that a ΣΔ ADC has over other architectures? Why is it the preferred choice for temperature sensors?

Also, on a related note, are there any continuous time implementations  for a ΔVbe based Temperature sensor? Put another way, can the noise and bandgap errors be within limits to get a ±2 degrees precision?

ywguo wrote on Jun 15^th, 2013, 9:40pm:


Hi Yawei,

Many thanks again for your response. The ±2 degrees precision is required over -40 to +95 ∘C. One point calibration is allowed.

I forgot to mention -- and it's completely my fault -- but the idea is to sense temperate in a remote location. That means I can't use 2 BJTs, only one (due to cost). This implies that I can't use Tuthill's switching scheme (which is patented anyway) and my ΔVbe will only increase by 200-300 uV per ∘K. That is, with a ratio of 20*I vs just I. I could increase the ratio, but can't really keep on increasing it to the point that the output voltage moves into mV.

I think, and I'm probably asking a rhetorical question here, that necessitates the use of some form of auto-zeroing or chopper stabilization to remove the amplifier offset. If any of these techniques is used, I don't think I can use a simple resistor-based non-inverting amplifier to amplify the signal.

The other thing is, with auto-zeroing, the thermal noise folds back into the band of interest. So the only option is chopping. And a switched-cap implementation to reduce the noise by increasing the size of the sampling capacitor.

Is there any other way to implement a remote temperature sensor using a ΔVbe scheme?

A couple of things. Bakker turned his thesis into a book. If it is the one I'm thinking of (which had a picture of his son on the cover) it could be very helpful.

http://books.google.com/books?id=TTw7udu3EqIC&pg=PA63&lpg=PA63&dq=bakker+thesis+...

Somewhere in there he had a clever trick on how if you put a temp co in the bandgap reference it will remove the curvature from the temperature measurement.

I do not see why subtracting a constant voltage would cause an error.

Check the Westwick and Gilbert patents in the references of the paper I wrote a paper a while back using a switched current to generate a reference with a single diode (http://web.mit.edu/Magic/Public/papers/01661769.pdf). It was a constant voltage and different than what you need, but you might be able to see one way to do it if you read between the lines. If not, at least glance at the references.

Delta-sigmas are used because they can be very accurate and the noise is oversampled. Temperature changes so slow you don't need a fast sample rate. You can also set up a DSM loop where you sample the PTAT signal, but subtract off a Vbe+PTAT voltage based on the same bipolar instead of creating a separate bandgap reference.