Some EE opinions please

[Incoherent]

I'm building a 12bit multichannel thermometer/controller using thermistors and a MAX186 amongst other things.
I need opinions on the analog side. The MAX186 datasheet will not tell me the input impedance of the analog inputs.
I want to use a voltage divider with one precision resistor and a thermistor between Vref and ground.
It is advantageous to use higher resistance (100K-1M) thermistors because there is less selfheating, but that leads to the question.
Would I need to buffer the inputs to prevent the ADC itself affecting the reading?
Thinking of a LM339 as a voltage follower, is this a stable solution? i.e. will it oscillate?
I'd like to avoid avoid extra circuitry.



Cheers

Incoherent

[Groth]

They don't give an input impedance because it doesn't have one in the traditional sense. It periodically charges a sample-and-hold capacitor: during a three clock tick sampling period, input impedance is very low ~5K; the rest of the time, impedance is effectivly infinite. So the net current drawn by the ADC inputs depends on the sample frequency, and the convertion accuracy depends the host clock and on higher current during the sampling period.

For your thermistors, bandwidth doesn't terribly matter. So the easiest with them would be to put a 10nF cap between the input and ground. The cap would supply the higher currents for the sampling phase, and the net current through your thermistor voltage divider could be kept low enough to not affect the results by cutting the sample rate to 10/s or less.

If you need to measure thing with high bandwidths, you'll need voltage followers on the ADC inputs.

[Incoherent]

Originally Posted by Groth

They don't give an input impedance because it doesn't have one in the traditional sense. It periodically charges a sample-and-hold capacitor:

Ah. Thank you Groth, an long unused section of my brain suddenly kicked back to life. I think it would probably be safer to use a voltage follower. My intention is to only sample at 1s but that might increase later. It'd let me not worry about the thermistor current. Although low component count is a priority, accuracy is more important.

[Groth]

Yeah, the voltage follower will allow the board as a whole to be more flexible, making it simpler to switch a channel to a different task.

Sounds like a interesting project.

[Incoherent]

Originally Posted by Groth

Sounds like a interesting project.

I hope so. I am setting up a poor mans die simulator using thermistors and commenced circuit designing when I found almost exactly what I had drawn up.Here but a different ADC. I will use the software and elements of the design. 8 analog channels (thermistors, flowrate, pressure), 4 digital inputs, and 4 output bits to control power level flowrate etc.
I like thermistors, they are easy and more sensitive than thermocouples, RTDs etc so I hope the repeatability will be enough for a good calibration. Will keep me occupied.
Damn I need a time machine.

[Groth]

That's a neat little kit they've got. Seems their pdf's are broken though (or my reader is).

Time machine?

[Incoherent]

Originally Posted by Groth

That's a neat little kit they've got. Seems their pdf's are broken though (or my reader is).

Time machine?

Simply that it'd allow me to do what I need to do simultaneously with what I want to do. Currently I am too busy so this little project needs to be on hold for a few weeks.

The PDFs are working for me.

[firtol88]

Originally Posted by Incoherent

...Here...

Nice find.

[dnkroz]

Hi.

Incoherent, you can also use integrated circuits like the LM35 or similars. It has a very linear answer and it is not necessary to have voltage stable.

The adaptation is relatively simple. I have a similar ADC almost finished.

[Incoherent]

Originally Posted by dnkroz

Hi.

Incoherent, you can also use integrated circuits like the LM35 or similars. It has a very linear answer and it is not necessary to have voltage stable.

The adaptation is relatively simple. I have a similar ADC almost finished.

Hi dnkroz
Thanks for the suggestion, I have considered it. The only problem with the LM35 type sensors is their size and shape. I want to insert the sensor into a small (~1mm) hole. Thermistors are available in a very small package.

[DeadEye]

Originally Posted by Incoherent

Hi dnkroz
Thanks for the suggestion, I have considered it. The only problem with the LM35 type sensors is their size and shape. I want to insert the sensor into a small (~1mm) hole. Thermistors are available in a very small package.

A few points that may/may not be useful:

You mention accuracy is important, the MAX186's internal reference depending on version can be as bad as +/- 80ppm. If using this chip ensure you get the best version. Much better external references are available.

Incorporating switchable gain instrumentation amplifiers into the design will allow various ranges to be monitored at better resolution.

Linearisation has to be incorporated in the software to handle the thermistor, if working over a shortish range then dependant on accuracy you are trying to achieve a small PRT, could be used which is almost linear over limited ranges.

Consider setting up a small current source (100uA) for powering of sensors, PRT, thermistors or as a resistor measuring device.

Hope this helps.

[Incoherent]

Originally Posted by DeadEye

A few points that may/may not be useful:

You mention accuracy is important, the MAX186's internal reference depending on version can be as bad as +/- 80ppm. If using this chip ensure you get the best version. Much better external references are available.

Incorporating switchable gain instrumentation amplifiers into the design will allow various ranges to be monitored at better resolution.

Linearisation has to be incorporated in the software to handle the thermistor, if working over a shortish range then dependant on accuracy you are trying to achieve a small PRT, could be used which is almost linear over limited ranges.

Consider setting up a small current source (100uA) for powering of sensors, PRT, thermistors or as a resistor measuring device.

Hope this helps.

Appreciate the input Deadeye.
I was planning to use a MAX6241 voltage reference via a OP484 type voltage follower buffer to feed both the ADC and the thermistors. Settling on 100K thermistors and 80K 0.1% precision resistors to minimise self heating and lead resistance effects, buffered by said OP484 voltage followers. You sound like you might be able to say if this is a reasonable approach or not.
Linearisation and calibration will be in software (Excel, heh)
Tell me more about these intrumentation amplifiers, I've been looking at the AMP02 but am not sure if I need/can even extract from noise the extra resolution. Plus I want to keep the single rail if possible.
I'm over 15 years beyond studying this stuff.

[DeadEye]

The approach you are looking at seems fine to me, you didnt explain what the whole design is for. If you are just making a basic temperature monitoring sytem covering a limited range only, then I would stick with what you have now. If you are designing a more flexible system for doing other things then generally you need to add optional gain to the circuit.
I know you realise that 4096mV is 1mv per step resolution, so it is easy to work out what you get on your thermistor in this application, however if great ranges are required then instrumentation amps start to offer the benefits. I preffered the Burr Brown (now TI) range due to their low power requirements.
Instrumentation amps offer benefits in precise gain, common mode rejection and ability to run from dual/single supplies. It is also possible to get the inputs/outputs to go to within 100mV of the rail supply.
Just one other comment, I take it you are running fairly short cables for the thermistor ?
If running longish cables then consider the use of constant current supply rather than a voltage divider.

[dnkroz]

I´am using the OP177 to amplify the Vout of the LM35. The obtained resolution is very near to 0.01ºC (0.0001V) with a ADC of 10Bit, 0-2.5V. Without the amplification the resolution is pproximately 0.06ºC

In the measurement of CPU temperature is a relativity problem because the measurement scale is reduced to 40ºC, but for the rest of measures (11 channels) of the system it would be sufficient.

The solution would be obvious to reduce the resolution in this channel (smaller amplification) or to reduce a fixed tension. For example -150mV (15ºC), it would increase the scale of measurement.

Here some images can be seen about my system, and this link NTC NTC2 if possible that interesting for you.

Still I have left tests to do. Sorry my bad english

[Incoherent]

Can anyone suggest a stable, simple 12v>5v voltage regulator circuit.
Currently I am leaning towards simply a LM317 with precision resistors. I am uncertain of the temperature stability of this and would like to lock it to the 4.096v voltage reference I am using as reference for ADC and thermistors.
I am anticipating that the 12v in will not be particularly stable, it will probably be powered from a hard drive connector initially. Any thoughts on whether power supply stability is even necessary in this case.

The whole circuit is based heavily on this but with a higher end analog side. I am intending to use the same software or a variant.

dnkroz, Deadeye, a belated thanks for your inputs, this thread dropped off my radar over the holidays.
I have decided not to use instrumentation amps at this point, the accuracy I can reasonably expect is going to make the 0.025 deg C resolution I am expecting quite adequate.

[bigben2k]

How about this:
http://www.analog.com/Analog_Root/pr...AD1584,00.html

or do you need some power to go with it?

Do you need it lower to cover the 317's voltage drop?

[Incoherent]

Originally Posted by bigben2k

How about this:
http://www.analog.com/Analog_Root/pr...AD1584,00.html

or do you need some power to go with it?

Do you need it lower to cover the 317's voltage drop?

Simpler than that Ben. I already have a Voltage reference, (MAX6241) what I need is a regulator circuit for the power supplies. I want this to be very stable, and I was wondering if it was possible to feed the voltage reference to the V adj pin of the 3TR in some manner where it was still regulating properly but was locked to the reference voltage. i.e if the reference drifts, so does the rest of the circuits power supplies, (to amps, sensors and ADC) thus cancelling any error this would generate. I realise the ADC will do it's thing relative to its ref V but I want to reduce the uncertainty.
Is this necessary? The voltage followers would (should) not be affected by Vcc fluctuations either but I am integrating a shunt with an amp for measuring current which might be.

[bigben2k]

Last I checked, a DAC's Vref should come from a seperate source.

I think you're trying to steady the supply voltages for the whole circuit, if I read you correctly. The diagram seems to have the 317 dictating output voltages which feeds Vsupply for all the other chips.

Last I checked, all these components work within a certain voltage range, i.e. 4.5 to 5.5 v. So you'd be looking to eliminate the transients. You can do that simply by adding caps right at the Vsupply and GND pins of each IC. An oscilloscope would give you a detailed picture. A standard PSU would have transients in the 1kHz to 10 kHz range (from the switching), and might have some around 60 Hz (from AC).

Does that help?

[Groth]

The LM317 will be sufficient. Add a 10 uF tantalum cap parallel to R8, a protection diode anti-parallel R7, another diode from LM317 terminal 2 -|>|- 3, make C3 a 0.1 uF ceramic, and make C4 a 1000 uF aluminum.

Couple that with good bypassing of the ADC's V+ (a 22 uF tant and a 2 nF ceramic in parallel, as close to V+ as possible) and you'll be fine.

Have you thought about feeding your thermistors with a larger voltage? Doesn't seem that you'll be getting the full benefit of your 12 bits/4.096 range.

I'd also add protect diodes to your MAX6241, pin 6 -|>|- 2 and 3 -|>|- 2.

[Incoherent]

Thanks for the bypassing/protection suggestions Groth. Gives me solid numbers for values. I will stick with the LM317.

Originally Posted by Groth

Have you thought about feeding your thermistors with a larger voltage? Doesn't seem that you'll be getting the full benefit of your 12 bits/4.096 range.

Yes I have. Two problems I see.
1. I would need a -ve supply to counterbalance the higher voltage and keep the operating level within the 0-4.096 range, if I wanted to keep the circuit as it is.
2. I'd like to keep the thermistor voltage tied to the same reference as the ADC, both the higher V+ and the V- would not be if I wanted to minimise component count.

A possible solution is this. I back track and DO use intrumentation amps.

Would this or a variation work?

[Groth]

Yeah, that variation would work. You'd definitely want intrumentation amps to avoid input bias current errors. Maybe use a single voltage divider with a voltage follower amp to supply all your channels.

If you want to stick to your present amps and have the thermistor feed voltages dependent on the Vref -- bring in a -12V line, then use an inverting amp with a bit of gain to derive an appropriate negative voltage from Vref, and a non-inverting with a bit of gain to get the postive.

[Incoherent]

Originally Posted by Groth

Yeah, that variation would work. You'd definitely want intrumentation amps to avoid input bias current errors. Maybe use a single voltage divider with a voltage follower amp to supply all your channels.

Good idea, I think I'll go with this.

Originally Posted by Groth

If you want to stick to your present amps and have the thermistor feed voltages dependent on the Vref -- bring in a -12V line, then use an inverting amp with a bit of gain to derive an appropriate negative voltage from Vref, and a non-inverting with a bit of gain to get the postive.

Another good, simple idea, you really are full of them aren't you. I like not having to use another supply though so...

[Groth]

Well. if you want to stick to single supply, you could use a charge pump to generate the small amounts of negative voltages you would need...

But, I do like the first variation and the instrumentation amps will give you better signal-to-noise, common-mode-rejection, etc. If you can, pick a low power single supply amp and run it off your Vref (may require a beefier voltage follower); that way the ADC inputs are guaranteed be to be with bounds.

Usually when someone calls me full of it, they're talking about something else.

[Incoherent]

Originally Posted by Groth

If you can, pick a low power single supply amp and run it off your Vref (may require a beefier voltage follower); that way the ADC inputs are guaranteed be to be with bounds.

Something like this is what it is turning into. A bit of a Hydra. Cut out one component, two others grow back.

Two high resolution (<0.005 deg) low range channels for water In/Out, 4 low resolution (<0.025 deg) for three sensor fluxblock and die. One ammeter channel and one voltmeter channel. I think I am going to skip the digital in/outs but we'll see.

What a mess. Now to see if I can transfer this onto a PCB.

Opinions/corrections please.

[Groth]

Hah, looks like you're designing a plate of spaghetti! What software are you doing the schematics with? They look great.

Generally looks good to me. What sort of the currents are you running through your heater? Might be better to move your current sensing shunt off board.

If you don't do the digital ins and outs yet, how about leaving yourself solder pads or a header that gives access to +5/ground/p-port lines. That way you could attach a daughter card later.