>
> MCP9800/1/2/3 2-Wire High-Accuracy Temperature Sensor
>
> - ±0.5°C (typ.) at +25°C
> - ±1°C (max.) from -10°C to +85°C
>
I dunno. 1 degree C doesn't seem very accurate to me. I mean, one
degree F (5/9 degree C) makes a difference between "comfortable"
and "too cold" in my home temperature...
(Does anyone make a sensor specifically aimed at human-range
temperatures, say 0.1 degree C over the range 10 to 30 C?)
> I dunno. 1 degree C doesn't seem very accurate to me. I mean, one
> degree F (5/9 degree C) makes a difference between "comfortable"
> and "too cold" in my home temperature...
I suspect most people would find it hard to detect that sort of
difference on an absolute basis. As a comparison eg between two rooms
it would be easier. You must be that most unusual of creatures - a
sensitive engineer. :-)
> (Does anyone make a sensor specifically aimed at human-range
> temperatures, say 0.1 degree C over the range 10 to 30 C?)
Platinum RTDs probably suffice but are expensive. (100 ohm at ?20C?
standardised platinum resistors).You could probably "roll your own"
for less than commercial.
You can do EXTREMELY well with some thermistors, sold expressly for
this purpose - probably as good as you want.
A gas thermometer can be extremely accurate - but you then need to
measure other parameters such as pressure.
The two-current method with a silicon diode is cheap, accurate and
little known. A briefish googling found nothing. / If you successively
apply two different currents to a diode the difference in voltage
drops is related to temperature. Within a given family of diodes the
method is independent of the particular diode being used.
Anyone else got any information on practical implementations of this
method?
>>MCP9800/1/2/3 2-Wire High-Accuracy Temperature Sensor
>>
>>- ±0.5°C (typ.) at +25°C
>>- ±1°C (max.) from -10°C to +85°C
>I dunno. 1 degree C doesn't seem very accurate to me. I mean, one
>degree F (5/9 degree C) makes a difference between "comfortable"
>and "too cold" in my home temperature...
>
>(Does anyone make a sensor specifically aimed at human-range
>temperatures, say 0.1 degree C over the range 10 to 30 C?)
I haven't seen one with those specs, but my students are using a Maxim DS1626 in their ROV and they tout a 0.5 degree C from 0 to 70C and overall operation from -55C to +125C. The ROV spent four hours in the tropical reef on Saturday at the Aquarium of the Pacific in Long Beach. The reef is kept at 78F degrees and alarms go off if it gets to 77F or 79F. The ROV was reading 25.5C (77.9F) all day. They have the sensor set for 11 bits and about a 375mS conversion time. They have an identical sensor on the processor board to monitor the inside temps of the control box. Worst case there was 93F.
Very easy sensor to interface with. We did the ICE calibration and also various temps up to 70C and were always within 0.25 of the centigrade thermometer.
Accuracy is as good as you want to pay for: ).15C / 0.03C / 0.001 C...
Quote:
Sensor manufacturers offer a wide range of sensors that
comply with BS1904 class B (DIN 43760): these sensors offer an
accuracy of ?0.3 C at 0 C. For increased accuracy, BS1904 class A
(?0.15 C) or tenth-DIN sensors (?0.03 C). Companies like Isotech can
provide standards with 0.001 C accuracy. Please note that these
accuracy specifications relate to the SENSOR ONLY: it is necessary to
add on any error in the measuring system as well.
>> - ±0.5°C (typ.) at +25°C
>> - ±1°C (max.) from -10°C to +85°C
>>
>I dunno. 1 degree C doesn't seem very accurate to me. I mean, one
>degree F (5/9 degree C) makes a difference between "comfortable"
>and "too cold" in my home temperature...
Those are as good as I have seen for an off-the-shelf device. Even the best
AD590 devices I can buy are +/-0.5C at 25C, and they cost a mint. Remember
that +/-1C is over a pretty wide range, and at about the centre of the range
you are looking at half that.
The only way to get better than that is to individually calibrate each
device, or go to something like a Pt100 or thermocouple sensor, and then you
need a reasonable amount of other electronics to get the accuracy.
I would suggest that a spot check at about three points on the curve, and
some software linearisation would do to improve a standard sensor like that
quoted.
>>MCP9800/1/2/3 2-Wire High-Accuracy Temperature Sensor
>>
>>- ±0.5°C (typ.) at +25°C
>>- ±1°C (max.) from -10°C to +85°C
>I dunno. 1 degree C doesn't seem very accurate to me. I mean, one
>degree F (5/9 degree C) makes a difference between "comfortable"
>and "too cold" in my home temperature...
Don't confuse resolution with precision..
>(Does anyone make a sensor specifically aimed at human-range
>temperatures, say 0.1 degree C over the range 10 to 30 C?)
>
>BillW
> >
> > MCP9800/1/2/3 2-Wire High-Accuracy Temperature Sensor
> >
> > - ±0.5°C (typ.) at +25°C
> > - ±1°C (max.) from -10°C to +85°C
> >
> I dunno. 1 degree C doesn't seem very accurate to me. I mean, one
> degree F (5/9 degree C) makes a difference between "comfortable"
> and "too cold" in my home temperature...
>
> (Does anyone make a sensor specifically aimed at human-range
> temperatures, say 0.1 degree C over the range 10 to 30 C?)
The Dallas 1-wire part DS18B20 can give .0625C resolution (1/16 degree
C).
For my home temp logger, I use an LM34 and a Vref+ on my ADC of 1.024V.
This gives me .1F (1mV) resolution (10 bit ADC) but a maximum temp of
only 102.3F with a minimum of 0F. I suppose that I could supply a Vref-
value and slide the whole scale up a few degrees or use an op-amp to
offset the output of the LM34, but my house should never be over 100F.
> The Dallas 1-wire part DS18B20 can give .0625C resolution (1/16 degree
> C).
>
> For my home temp logger, I use an LM34 and a Vref+ on my ADC of 1.024V.
> This gives me .1F (1mV) resolution (10 bit ADC) but a maximum temp of
> only 102.3F with a minimum of 0F. I suppose that I could supply a Vref-
> value and slide the whole scale up a few degrees or use an op-amp to
> offset the output of the LM34, but my house should never be over 100F.
Why not use an opamp to scale the LM34 output 2x. That way the
circuit would output 20mV/degree, and if using only 8 bits of the ADC,
the number you'd get from it would be the acutual temperature,
accurate to within one degree (rounding error).
Here's an example:
LM34 output = .72V (72 degrees)
scale 2x = 1.44V
ADC resolution (5V Vref, use only ADRESL value) = 5/256=0.01953125
(round to .020mV/bit)
1.44V/.020V/bit=72<--value in ADRESL
Are you sure? 1 degree C is negligible IMHO. Perhaps there are other factors
(such as humidity involved)? And I feel certain that different parts of your
home may have variations of way more than 1 degree.
Cheers,
-Neil.
On Tuesday 22 March 2005 12:04 am, William "Chops" Westfield scribbled:
> I dunno. 1 degree C doesn't seem very accurate to me. I mean, one
> degree F (5/9 degree C) makes a difference between "comfortable"
> and "too cold" in my home temperature...
>
> (Does anyone make a sensor specifically aimed at human-range
> temperatures, say 0.1 degree C over the range 10 to 30 C?)
>
> BillW
On Tuesday 22 March 2005 04:16 am, michael brown scribbled:
> For my home temp logger, I use an LM34 and a Vref+ on my ADC of 1.024V.
> This gives me .1F (1mV) resolution (10 bit ADC) but a maximum temp of
> only 102.3F with a minimum of 0F. I suppose that I could supply a Vref-
> value and slide the whole scale up a few degrees or use an op-amp to
> offset the output of the LM34, but my house should never be over 100F.
I was curious how you use a 1.024V reference (I'm assuming this is on a PIC),
as PICs (16F872 as example) require a minimum Vref of about 2V. Or are you
measuring the reference itself as another input and comparing to the
temp-sensor input?
Shouldn't the higher current produce the higher voltage drop??
As the forward voltage drop is ~exponential with current, I think you may
need a 10:1 ratio for the 2 currents to get a reliable estimate.
One suggestion I was going to make re the temp sensor is to connect a
couple of transistors in darlington configuration and then the "base" to
"collector". This produces a quite flat voltage/current charecteristic, a
voltage drop of about 1.2V and twice the temperature coeficient of a single
diode.
Richard P
I have been experimenting with sending 2 currents through a diode but find
no real change in the voltage drop difference as temp changes.
>>I dunno. 1 degree C doesn't seem very accurate to me. I mean, one
>>degree F (5/9 degree C) makes a difference between "comfortable"
>>and "too cold" in my home temperature...
>
> Don't confuse resolution with precision..
He doesn't appear to have been.
The above sensation relies on both precision and resolution.
> > I have been experimenting with sending 2 currents through a diode but find
> > no real change in the voltage drop difference as temp changes.
> >
> > @67F 1mA = .675volts 2mA = .589volts = .089volts diff.
> > @77F 1mA = .611volts 2mA = .521volts = .090volts diff.
>
> Shouldn't the higher current produce the higher voltage drop??
> As the forward voltage drop is ~exponential with current, I think you may
> need a 10:1 ratio for the 2 currents to get a reliable estimate.
The other problem with this type of measurement is that it relies on a
difference. Once you use your measurement incertainties in the
calculations, you see the result incertainty being quite high mostly.
> On Tue, 22 Mar 2005 04:16:29 -0600, michael brown
> <spam-mespam_OUThouston.rr.com> wrote:
>
> > The Dallas 1-wire part DS18B20 can give .0625C resolution (1/16
degree
> > C).
> >
> > For my home temp logger, I use an LM34 and a Vref+ on my ADC of
1.024V.
> > This gives me .1F (1mV) resolution (10 bit ADC) but a maximum temp
of
> > only 102.3F with a minimum of 0F. I suppose that I could supply a
Vref-
> > value and slide the whole scale up a few degrees or use an op-amp to
> > offset the output of the LM34, but my house should never be over
100F.
>
> Why not use an opamp to scale the LM34 output 2x. That way the
Because I wanted 1/10 degree resolution. :-) I didn't care about
losing some top end of the scale. Even though any given temp may be off
by ~1 degree, the relative measurements are very accurate.
By taking samples at 1 minute or less (I've been using 30 seconds) and
plotting the readings, you can tell some interesting things about a
house. Of course you can easily see when the AC/heat kicks on and off.
You can even clearly see when people just walk into a room for a moment.
A relatively unused room upstairs in my home (no thermostat in the room,
only one downstairs) follows a 24 hour sine wave of temperature
fluctuation during AC season. The average temperature swings about 2-3
degrees over the course of the day/nite. Of course this all depends
upon outside temperature/cloud and sun conditions/winds. I've only done
a little bit of logging so far, but I look forward to getting a whole
years worth of data on all the rooms plus the attic and outside temps.
> circuit would output 20mV/degree, and if using only 8 bits of the ADC,
> the number you'd get from it would be the acutual temperature,
> accurate to within one degree (rounding error).
>
> Here's an example:
> LM34 output = .72V (72 degrees)
> scale 2x = 1.44V
> ADC resolution (5V Vref, use only ADRESL value) = 5/256=0.01953125
> (round to .020mV/bit)
> 1.44V/.020V/bit=72<--value in ADRESL
That's clever, but I want to keep my 1/10 degree resolution. I would
like to raise my upper temp limit a bit though. I've been thinking
about switching to a sensor that has a PWM output. Another thought is
to go with I2C since I already have a serial EEPROM and DS1307 RTC in
the circuit. Any recommendations?
> Because I wanted 1/10 degree resolution. :-) I didn't care about
> losing some top end of the scale. Even though any given temp may be off
> by ~1 degree, the relative measurements are very accurate.
There are real advantages to using a thermistor. First off, they are
ratiometric, so you don't have competing references. They are also quite
inexpensive.
All you need to use is one opamp to scale them - .02 degree resolution is
nothing. Your span is reduced, but you said you didn't need much. Use
another analog input to read the thermistor directly if you need to
measure out-of-range temperatures.
I'd also like to speak in favor of thermistors. I'm working on an
application with 0.1 degree accuracy in the area immediately around 100F
(probably plus or minus 20 degrees F). I used a Maxim Data Aquisition
System chip that includes a 16 bit delta sigma A/D, a D/A, and a Real Time
Clock. It has its own reference on it. I put a 0.1% resistor between the
reference and the top of the thermistor, then connected the junction to
the analog input. The Maxim DAS chip has an SPI interface that talks to
the PIC. The thermistor manufacturer provided a table of temperatures
versus resistance. I used a spreadsheet to convert the resistances in the
table to actual A/D counts based on the resistance and the 0.1% resistor.
This table was then put in an array in the PIC. The PIC then uses linear
interpolation between the values supplied by the manufacturer.
>> Because I wanted 1/10 degree resolution. :-) I didn't care about
>> losing some top end of the scale. Even though any given temp may be off
>> by ~1 degree, the relative measurements are very accurate.
>
> There are real advantages to using a thermistor. First off, they are
> ratiometric, so you don't have competing references. They are also quite
> inexpensive.
>
> All you need to use is one opamp to scale them - .02 degree resolution is
> nothing. Your span is reduced, but you said you didn't need much. Use
> another analog input to read the thermistor directly if you need to
> measure out-of-range temperatures.
>
> Cheerful regards,
> Bob
> On Tuesday 22 March 2005 04:16 am, michael brown scribbled:
> > For my home temp logger, I use an LM34 and a Vref+ on my ADC of
1.024V.
> > This gives me .1F (1mV) resolution (10 bit ADC) but a maximum temp
of
> > only 102.3F with a minimum of 0F. I suppose that I could supply a
Vref-
> > value and slide the whole scale up a few degrees or use an op-amp to
> > offset the output of the LM34, but my house should never be over
100F.
>
>
> I was curious how you use a 1.024V reference (I'm assuming this is on
a PIC),
Yes it is a PIC 16F88
> as PICs (16F872 as example) require a minimum Vref of about 2V. Or
are you
> measuring the reference itself as another input and comparing to the
> temp-sensor input?
Nah, I'm just carelessly violating the specs. ;-) I'm usually pretty
good about reading the datasheet, but I never looked that up. Since
I've never used the ADC before, it didn't even occur to me that there
would be a minimum voltage for Vref+, though in retrospect it seems
perfectly sensible that there is. The actual spec for the Vref+ minimum
is (Vcc-2.5V), so I'm way outside the bounds at 1.024V using a 5V Vcc.
It seems to work very well, but I did notice that I needed to use lots
of bypassing and to let everything acquiesce before converting a sample
in order to get consistent readings. That may or may not be relevant to
me being out of specs. I suspect it probably is.
mac.com>
