Resistor values to bias an opamp

[robca]

Hello,

I'm working on a very low power device, and I need to read 7 FSRs into an ADC. The front end is the circuit below, and currently I'm using two 1MOhm resistors to create the VREF/2 voltage (voltage divider between VREF and GND). The opamp used is a TSU114, and I'm using two of those (4 opamps each, with an unused ooamp). A prototype works just fine.

I'm really quite ignorant when it comes to analog circuits, so I'm wondering if I can use just 2 resistors in total for all 7 opamps (i.e. a single voltage divider providing vref/2 for all. I know I can use a voltage reference IC, but trying to minimize the amount of active devices to minimize sourcing risks like the ones we saw during the pandemic). I guess it all depends on the type of load the positive input of the opamp as seen by the voltage divider. If I need to use a voltage divider for each opamp input, can I use resistor values higher then 1MOhm, say 2,2MOhm or even 4.7MOhm. With two 1MOhm. the total current used by the 14 resistors is ~6.3uA, which is a non-negligible power compared to the rest of the circuit (considering it's always used, even when not doing ADC conversions)

What's the best way to set the bias on 7 opamps without using a voltage reference IC?

[BillyO]

Given the input current of the TSU114 is in the pA range you should be fine to use just 2 resistors.  5-10Mohm might be a good value.

[MasterT]

10 Mohm is o'k, but don't forget to put a cap at voltage divider output to ground, 0.1-10 uF to: minimize noise, interference between op-amps, RF or electrical networks.

[Jwillis]

Found this if it helps. http://fab.cba.mit.edu/classes/863.16/section.Architecture/people/Ge/bias_opamp.pdf

[Terry Bites]

The high value ressitors are fine for setting the DC operating point but they create a high AC impedance to ground. Bypass the lower resistor in the divider with a cap. Say 100n+10u.
Note that an additional offset is created, Vos=Ibias*R. Check that's in spec for your circuit. A TL431 is a handy reference and they only cost about $0.30. On Aliexpress you can get a 100 for $5! Worth as punt.

[robca]

First of all, thanks everyone for the replies and the additional reference material.

I'm trying to avoid adding active components. Yes, a TL431 is cheap, but if there's another 2021-like shortage, it's another potentially hard-to-source component to find (this board is a prototype for a small company which might become a product... and if you wonder why a clueless person like me is doing it, well, I'm mostly a digital/firmware guy helping friends)

Also, the device needs 7 channels, so I have an unused opamp to ensure is in low power mode (using two quad packages makes the most sense, compared to a quad, dual and single package).

It just so happens that the recommended way to prevent an unused opamp from using excessive power is a unity gain voltage follower with a voltage roughly of VCC/2. So if I use the voltage follower biasing from the document the @Jwillis suggested (https://fab.cba.mit.edu/classes/863.16/section.Architecture/people/Ge/bias_opamp.pdf) with R1=R2, I can kill two birds with a stone: the unused opamp is in ideal conditions, and I get my VCC/2 bias for all 7 opamps.

Even if the voltage bias were to be slightly off VCC/2, it won't matter as much because all 7 channels would be identical and I'll have to calibrate each channel in software anyway (but thanks for pointing that out)

Something like the circuit below with R1=R2=4.7MOhm for the reference voltage (including the bypass capacitors) and Vbias connected to the positive input of all 7 channel opamps should get me what I need

[Kleinstein]

The buffer for VS/2 is OK, just leave out the capacitor at the output.

The TL431 makes no sense in this context as the measurement is ratiometric and thus does not need a stable voltage, but wants all relative to the same voltage, like the supply. This would even work if the supply is not stable, but drifty, like directly from a battery.

[robca]

The buffer for VS/2 is OK, just leave out the capacitor at the output.

The TL431 makes no sense in this context as the measurement is ratiometric and thus does not need a stable voltage, but wants all relative to the same voltage, like the supply. This would even work if the supply is not stable, but drifty, like directly from a battery.

Thanks! to help me understand, why are you suggesting to remove the capacitor on the output? That was taken from an MIT paper, so from people who understand electronics far better than I do.

[Manul]

10 Mohm is o'k, but don't forget to put a cap at voltage divider output to ground, 0.1-10 uF to: minimize noise, interference between op-amps, RF or electrical networks.

Beware that some capacitors have significant leakage currents, especially electrolytic type. High density ceramic ones may also have leakage. Film capacitors are usually best.

[Kleinstein]

OP-amps usually don't like capacitive load at the output and may start oscillating with to much capacity. In this case the load to the buffered Vs/2 level is rather light and there is no need to buffer fast current spikes.

For the capacitor at the divider the leakage and DA can be an  issue that can slow down the settling a the start. With large resistors one could get away with a relatively small capacitor (e.g. 1-10 nF) and than use C0G ceramic. While X7R ceramics have lots of DA and other nasty effects, C0G ceramic is comparable to be better film capacitors (PP type), but get expensive with more than some 10 nF.

[MrAl]

Hello,

I'm working on a very low power device, and I need to read 7 FSRs into an ADC. The front end is the circuit below, and currently I'm using two 1MOhm resistors to create the VREF/2 voltage (voltage divider between VREF and GND). The opamp used is a TSU114, and I'm using two of those (4 opamps each, with an unused ooamp). A prototype works just fine.

I'm really quite ignorant when it comes to analog circuits, so I'm wondering if I can use just 2 resistors in total for all 7 opamps (i.e. a single voltage divider providing vref/2 for all. I know I can use a voltage reference IC, but trying to minimize the amount of active devices to minimize sourcing risks like the ones we saw during the pandemic). I guess it all depends on the type of load the positive input of the opamp as seen by the voltage divider. If I need to use a voltage divider for each opamp input, can I use resistor values higher then 1MOhm, say 2,2MOhm or even 4.7MOhm. With two 1MOhm. the total current used by the 14 resistors is ~6.3uA, which is a non-negligible power compared to the rest of the circuit (considering it's always used, even when not doing ADC conversions)

What's the best way to set the bias on 7 opamps without using a voltage reference IC?



(Attachment Link)

Hi,

What are you powering this circuit with, regular batteries or a mouse on an exercise wheel 

I am asking because using regular AAA batteries the current using 1M resistors is so low that the circuit would run for over 2 years even with 7 such circuits.  Maybe the op amps use a lot of power?

As you get up above 100k noise starts to creep in and could cause problems because it is going to be wideband which cannot be filtered out using passive components or even linear analog circuitry.  1M resistors might be ok though, or if you intend to do some oversampling then the noise might actually help.

[Kleinstein]

It looks like the sensors have a resistance in the Mohm range. So it is natural to use comparable range resistors also in the rest of the circuit. The low power OP's (the TSU114 takes ~1 µA per amplifier) are a bit noisey already, but good enough.  There are batteries smaller than AAA.

If one really wants super low power one could consider using the sensors to charge a capacitor and get way without extra OP-amps and activate the sensors (use additional IO ports) only when actually needed. I depends on the µC if this is feasable.

[robca]

What are you powering this circuit with, regular batteries or a mouse on an exercise wheel 

I am asking because using regular AAA batteries the current using 1M resistors is so low that the circuit would run for over 2 years even with 7 such circuits.  Maybe the op amps use a lot of power?

As you get up above 100k noise starts to creep in and could cause problems because it is going to be wideband which cannot be filtered out using passive components or even linear analog circuitry.  1M resistors might be ok though, or if you intend to do some oversampling then the noise might actually help.

Good point on the noise, which definitely is a consideration. We are still debating if using 1M or 2.2M resistors on the single divider of the new circuit.

I wish we were dealing something as big as a AAA battery The device is powered by a 50 or 70mAh battery, and has 2 states: active when transmitting over BLE and using ~3mA, then a sleep mode where the processor and wake up IMU are using a combined 9uA. The goal is to be able to spend most of the time in sleep mode (weeks) with sporadic transmission. Using 7 voltage dividers, almost doubles the sleep power for no added value. We are considering also an electronic switch to power off part of the circuit when it sleep mode, but that adds complexity

As per @Kleinstein point: sampling rate is pretty high (100Hz) and reading at the equivalent rate of 700MHz the sensors would not be feasible with our processor (which is also handling BLE). Not to mention that the FSR range from MOhms to kOhms, and I'm not sure a capacitor charge circuit would be capable of handing the wide range. Moreover, one of the nice side effects of using an opamp with an exponential curve is that it almost cancels out the exponential response of the FSR. Basically, the output of the opamp is an almost linear function of the force applied on the FSR (we still need to build a calibration curve, but we have better average resolution across the whole range of forces)

[Kleinstein]

Getting a nearly linear response is of cause nice.
The way with the capacitor can also use the µC internal ADC. By adjusting the delay from activating the sensor to reading the capacitors voltage one can actually get additional dynamic range.

In the version with the OP-amps one may also be able to deactivate the sensors by moving the non inverting inputs of the OP-amps from VS/2 to VS and this way save on the current through the sensors. Another possibly way would be to have the VS to the sensors through a µC pin and let it fload for off.

[robca]

In the version with the OP-amps one may also be able to deactivate the sensors by moving the non inverting inputs of the OP-amps from VS/2 to VS and this way save on the current through the sensors. Another possibly way would be to have the VS to the sensors through a µC pin and let it fload for off.

Interesting ideas... thanks!

[MrAl]

What are you powering this circuit with, regular batteries or a mouse on an exercise wheel 

I am asking because using regular AAA batteries the current using 1M resistors is so low that the circuit would run for over 2 years even with 7 such circuits.  Maybe the op amps use a lot of power?

As you get up above 100k noise starts to creep in and could cause problems because it is going to be wideband which cannot be filtered out using passive components or even linear analog circuitry.  1M resistors might be ok though, or if you intend to do some oversampling then the noise might actually help.

Good point on the noise, which definitely is a consideration. We are still debating if using 1M or 2.2M resistors on the single divider of the new circuit.

I wish we were dealing something as big as a AAA battery The device is powered by a 50 or 70mAh battery, and has 2 states: active when transmitting over BLE and using ~3mA, then a sleep mode where the processor and wake up IMU are using a combined 9uA. The goal is to be able to spend most of the time in sleep mode (weeks) with sporadic transmission. Using 7 voltage dividers, almost doubles the sleep power for no added value. We are considering also an electronic switch to power off part of the circuit when it sleep mode, but that adds complexity

As per @Kleinstein point: sampling rate is pretty high (100Hz) and reading at the equivalent rate of 700MHz the sensors would not be feasible with our processor (which is also handling BLE). Not to mention that the FSR range from MOhms to kOhms, and I'm not sure a capacitor charge circuit would be capable of handing the wide range. Moreover, one of the nice side effects of using an opamp with an exponential curve is that it almost cancels out the exponential response of the FSR. Basically, the output of the opamp is an almost linear function of the force applied on the FSR (we still need to build a calibration curve, but we have better average resolution across the whole range of forces)

Oh that is a small battery that's for sure.  That could be a major problem depending on how long you want the battery to last.
To get a 50mAh battery to last for a year you'd need to keep the average current down to about 6ua which is probably just too small.
If you add another 6ua, then you get roughly a half of a year, which may still be acceptable.
If we accept 7ua, then we'd need to keep the current for each set down to just 1ua, which would take a total divider resistance of 12.6M which means each resistor would be 6.3M.
Maybe you should look into oversampling, but then again that will take more processor time and at 3ma the time spent in the non-sleep mode has to be very short.  Microcontrollers made these days are pretty amazing that they can do this.
Yes, there is the possibility of turning off parts of the circuit that are not needed during the sleep time.  That would lower the current demand down to almost just the processor non-sleep current draw for the short time it is on.

What else we would need to know is how fast can the ADC take an analog sample and convert it.

[robca]

Oh that is a small battery that's for sure.  That could be a major problem depending on how long you want the battery to last.
To get a 50mAh battery to last for a year you'd need to keep the average current down to about 6ua which is probably just too small.
If you add another 6ua, then you get roughly a half of a year, which may still be acceptable.
If we accept 7ua, then we'd need to keep the current for each set down to just 1ua, which would take a total divider resistance of 12.6M which means each resistor would be 6.3M.
Maybe you should look into oversampling, but then again that will take more processor time and at 3ma the time spent in the non-sleep mode has to be very short.  Microcontrollers made these days are pretty amazing that they can do this.
Yes, there is the possibility of turning off parts of the circuit that are not needed during the sleep time.  That would lower the current demand down to almost just the processor non-sleep current draw for the short time it is on.

What else we would need to know is how fast can the ADC take an analog sample and convert it.

I don't need a year battery. Just not having to recharge frequently.

And we are down to a single divider, not 7 anymore (using the unused opamp in one of the quad packages as a voltage follower with VCC/2). We are currently on budget power-wise, thanks everyone for the input

[Kleinstein]

Depending on the time available from µC wake up to getting the readings it could be worth switching the ref. Voltage for the amplifier stages from Vs/2 (or a little more) to VS to disable the sensors. This would reduce the sensor current when not actually used. It however takes some time (some 1 ms) to bring the sensors up. For enough speed it may need a little smaller resistors at the divider (and no capacitor), but only active for a shorter time.

[robca]

Depending on the time available from µC wake up to getting the readings it could be worth switching the ref. Voltage for the amplifier stages from Vs/2 (or a little more) to VS to disable the sensors. This would reduce the sensor current when not actually used. It however takes some time (some 1 ms) to bring the sensors up. For enough speed it may need a little smaller resistors at the divider (and no capacitor), but only active for a shorter time.

Well, when the device is not being used, all the FSR are open, so they use no current. It's a wearable, and we use the IMU to wake up when on the user body. When not on the body, the FSR are pretty much open.

Or am I missing what you are suggesting?

[Kleinstein]

When the sensors are high impedance when open, it would not make much difference turning the sensors / amplfiiers off. It would mainly make a difference if there are longer phases with the sensors in a rel. low resistance state, where there is no measurement. The turn off option is low effort (use IO port instead of GND for the bias divider), but it depends on the details (free IO pins available ?, the FB resistor at the OP-amps, how high is the resistance if not in use, possible low R from abnormal unused state) if it is worth it.

A 100 SPS reading rate is already relative fast and possibly not worth / possible to turn off in between.

It may make sense to do the sampling at 2 x the mains frequency (e.g. 120 SPS in the US) and than average 2 readings. This could suppress mains hum as a possible source of interference.

[MrAl]

Oh that is a small battery that's for sure.  That could be a major problem depending on how long you want the battery to last.
To get a 50mAh battery to last for a year you'd need to keep the average current down to about 6ua which is probably just too small.
If you add another 6ua, then you get roughly a half of a year, which may still be acceptable.
If we accept 7ua, then we'd need to keep the current for each set down to just 1ua, which would take a total divider resistance of 12.6M which means each resistor would be 6.3M.
Maybe you should look into oversampling, but then again that will take more processor time and at 3ma the time spent in the non-sleep mode has to be very short.  Microcontrollers made these days are pretty amazing that they can do this.
Yes, there is the possibility of turning off parts of the circuit that are not needed during the sleep time.  That would lower the current demand down to almost just the processor non-sleep current draw for the short time it is on.

What else we would need to know is how fast can the ADC take an analog sample and convert it.

I don't need a year battery. Just not having to recharge frequently.

And we are down to a single divider, not 7 anymore (using the unused opamp in one of the quad packages as a voltage follower with VCC/2). We are currently on budget power-wise, thanks everyone for the input

Hello again,

I was wondering why you just don't use the SAME voltage divider for all of the op amps.  Why was there a need for seven in the first place.

Aside from that, I thought of the perfect solution.  Buy an 8 ampere-hour lead acid battery and strap that to the user's back (ha ha).
I take it you are currently using one of those small cylindrical 12v batteries used for remotes and stuff.
You did mention something about charging though.  Could you incorporate some sort of solar charger?
There are also "walking" chargers now, where the charging is done as the person walks, the movement causes the motion required for a electromechanical device to charge the battery.  Not sure if that would supply enough power though you'd have to check if this seems like an interesting option.

[robca]

Hello again,

I was wondering why you just don't use the SAME voltage divider for all of the op amps.  Why was there a need for seven in the first place.

As I explained in the first post, I wanted to use one divider for all 7, but I had read elsewhere that it might cause problems in certain cases. So I started with the basic circuit I knew worked, and cloned it 7 times as a starting point. Then I asked how to improve it. And while doing so, I realized that I could kill two birds with a stone, using the "unused" opamp to generate the bias voltage. So that's solved

Aside from that, I thought of the perfect solution.  Buy an 8 ampere-hour lead acid battery and strap that to the user's back (ha ha).
I take it you are currently using one of those small cylindrical 12v batteries used for remotes and stuff.
You did mention something about charging though.  Could you incorporate some sort of solar charger?
There are also "walking" chargers now, where the charging is done as the person walks, the movement causes the motion required for a electromechanical device to charge the battery.  Not sure if that would supply enough power though you'd have to check if this seems like an interesting option.

The device, like many wearables, uses a 50-70mAh lipo cell (and a charging circuit from USB). Solar chargers do not work for wearables, and add cost, complexity and points of failure.

Motion harvesting chargers are really not a mature technology, and whatever is available won't generate enough power to be worth the hassle of adding complexity, cost and yet another point of failure. I'm really bad at analog circuitry, but really familiar with wearables and low power devices (including devices with >1 year sleep time). There's really nothing better than a good quality lipo of the right size, and a proper charging plan/schedule.

[MrAl]

Hello again,

I was wondering why you just don't use the SAME voltage divider for all of the op amps.  Why was there a need for seven in the first place.

As I explained in the first post, I wanted to use one divider for all 7, but I had read elsewhere that it might cause problems in certain cases. So I started with the basic circuit I knew worked, and cloned it 7 times as a starting point. Then I asked how to improve it. And while doing so, I realized that I could kill two birds with a stone, using the "unused" opamp to generate the bias voltage. So that's solved

Aside from that, I thought of the perfect solution.  Buy an 8 ampere-hour lead acid battery and strap that to the user's back (ha ha).
I take it you are currently using one of those small cylindrical 12v batteries used for remotes and stuff.
You did mention something about charging though.  Could you incorporate some sort of solar charger?
There are also "walking" chargers now, where the charging is done as the person walks, the movement causes the motion required for a electromechanical device to charge the battery.  Not sure if that would supply enough power though you'd have to check if this seems like an interesting option.

The device, like many wearables, uses a 50-70mAh lipo cell (and a charging circuit from USB). Solar chargers do not work for wearables, and add cost, complexity and points of failure.

Motion harvesting chargers are really not a mature technology, and whatever is available won't generate enough power to be worth the hassle of adding complexity, cost and yet another point of failure. I'm really bad at analog circuitry, but really familiar with wearables and low power devices (including devices with >1 year sleep time). There's really nothing better than a good quality lipo of the right size, and a proper charging plan/schedule.

Hi,

Sounds great, good luck with the project.