High precission I->V converter

[xorly]

Hello all forum members!

I have come to problem, where my knowledge is not sufficient. Analog circuits are not my best friend after all.

I am searching best way how to convert current (+-7.5mA) to voltage (+-3V), DC~300Hz. Converter is meant for INN-204 accelerometer (parameters http://www.sussexsensors.co.uk/media/16f80079a1191f55ffff8388d4355564.pdf).

The problem is, I need to get to 5ug resolution as close as possible. Output is approx. 1.5mA/g. If I am calculating correctly, I need to recognize 1.5/(1/0.000005)= 7.5nA changes in +-7.5mA range. This sounds almost insane to me.
Enemy seems to be leakage current, noise (especially on DC), varying input offset...? Leakage current can be decreased by active guard ring -> Op Aap in circuit. Offset can be brought down by chopper op amps. Noise is present everywhere, but where is least?

I thought about 2 possible circuits:
- transimpedance amplifier (TIA)
- sensing resistor with voltage follower to create driven guard ring and to create stiff enough signal for further processing

Which one can achieve better results? (why?)
Are there any better approaches?

[Christe4nM]

That's quite a dynamic range you need.
Is there a specific reason you need that kind of resolution and/or precision / dynamic range?
In any case the data sheet also specifies a tempo of "<100 ug/°C" and a one year stability of "< 1000 ug/°C". So unless they can give you a more accurate and guaranteed value for those, you need to take those values as the best the manufacturer guarantees. In that case I'm not sure if that 5ug resolution still means anything useful?

[xorly]

Thanks for quick reply!
This is for inertial measurement unit, which will be used in air (airplane).
Temperature stability is not such problem. Temperature correction was tested on less sensitive design successfully. In worst case, whole unit will be placed in thermostatic chamber. Time stability will be resolved by periodic calibration.

I agree, that 5ug resolution is unrealistic with such stability parameters. I just need to get as close as possible. Sensor stability and other errors should become limit and not converter itself. Old design (which I need to rework, bad PCB) has noise equal to 1mg on output. I want to get as much as possible from new design.

[moffy]

When I worked with Inertial Navigation Systems(INS) the accelerometers fed their current into a capacitor which was then discharged by a precision current source/pulse. Sort of an I->F converter. It was very accurate. In fact the INS had gravity maps embedded to compensate for local gravity variations.

[xorly]

Wow, this sounds really interesting. Can you share more details? Kinda sounds like integrating ADC implementation.

[moffy]

Sorry, don't have the details now and even if I did I couldn't share then as they are proprietary. But yes an integrating A/D is a good comparison. Also V/F converters except you only need the I/F part, makes it a little simpler. Jim Williams designed a super linear 0 to 1MHz V/F converter, I think it is in the Linear Technology app notes. I also seem to remember that integrating the force(accelerometer current) is much more accurate.

[mycroft]

I think it is a Bob Pease application note for LM331. Take a look at http://www.ti.com/lit/an/snoa594b/snoa594b.pdf

[Alex Nikitin]

Let's do some simple reality check calculations first. 10mA maximum current, 10V output from the TIA, 1kohm feedback resistance. Self-noise of 1K resistor in a 400Hz bandwidth at room temperature is roughly ~80pA RMS. That is your physical limit of the dynamic range. You will need a decent precision opamp with low bias and noise currents, a very good DC stability (<5uV drift and LF noise) and a low noise voltage (below 5nV/rtHz). The opamp would increase the noise somewhat but <200pA RMS noise in 400Hz BW should be possible to achieve without much trickery.

Cheers

Alex

P.S. I suspect the resolution would be limited by the noise from the sensor, not by the TIA or the ADC after it.

[Christe4nM]

Wow that I->F idea is really cool. Can this also keep up with the ~300Hz bandwidth the OP needs?

(I've been banging my head trying to think of anything that might help the OP. But I got stuck with that big dynamic range and my own limited design experience. If you went current->voltage->ADC then you'd need a very good 20+ bits ADC measuring down to a few µV's resolution. That would most likely introduce all kinds of new difficulties with noise, linearity etc.)

[bobaruni]

Integrator does away with most of the resistor noise, just need a very quiet op amp and ADC.
This is just an example or you could roll your own version using state of the art low noise (0.1 -> 10Hz) op amps.
http://www.ti.com/lit/ds/sbfs009/sbfs009.pdf

[David Hess]

Wow that I->F idea is really cool. Can this also keep up with the ~300Hz bandwidth the OP needs?

(I've been banging my head trying to think of anything that might help the OP. But I got stuck with that big dynamic range and my own limited design experience. If you went current->voltage->ADC then you'd need a very good 20+ bits ADC measuring down to a few µV's resolution. That would most likely introduce all kinds of new difficulties with noise, linearity etc.)

Most voltage to frequency converters are ultimately current to frequency converters; their input circuit operates like an inverting amplifier so just remove the input resistor.  And they are a type of integrating converter with their integration time dependent on how long counts are accumulated; this produces the same sin(x)/x frequency response that any other integrating ADC has.  They can also have enormous dynamic range; Jim William's King Kong VFC operated from 1 Hz to 100 MHz which is 160 dB.

Linear Technology application note 14 is a good place to start as it covers various designs in one spot and includes a section at the end describing the major VFC topologies.  None of these converters achieve 1ppm linearity but if that is a design requirement, then you may be in trouble no matter how you do it.

Alex Nikitin covered the dynamic range of a transimpedance amplifier.  For more details on transimpedance amplifier performance including noise and bandwidth, check out various photodiode application notes where they are very commonly used.

7.5nA out of 7.5mA is 1ppm or about 20 bits of resolution.  That seems achievable to me but will require careful design and I agree that the sensor is not going to support this level of precision.  An integrating type of converter (current-to-voltage, delta-sigma, slope-integrating, whatever) will help a lot by inherently limiting noise.

If an integrated transimpedance amplifier is used, unload the output with a buffer because the heat produced by driving 7.5mA full scale with a precision operational amplifier will compromise its precision by lowering its open loop gain among other things.

[xorly]

Integrating voltage into capacitor seems to be the way. Existing integrating ICs seems to be made for APDs and thus, they have too small input range. I'd like to sample approximately 1ksps. I can imagine several analog switches for integrating part. Output will be processed by 200MHz ARM MCU. Time measurement resolution 5ns does not seem to be enough to measure discharging time. So ?? ADC?

[Marco]

Which one can achieve better results? (why?)

What is the compliance and impedance of the current source? That product brief disguised as a datasheet offers no clue.

Making a TIA when they've dedicated circuitry to make a really good current source doesn't make a whole lot of sense.

PS. can you really get a free lunch by integrating into a capacitor? Doesn't a switched capacitor just act like a resistor again?

PPS. the capacitor reset gives you offset correction for free though, so that's nice.

[xorly]

What is the compliance and impedance of the current source? That product brief disguised as a datasheet offers no clue.

Unfortunately, there is no decent datasheet. The only other paper I have is calibration list. The only info about output is in attachment. R_L is between 100R and 130R.

Doesn't a switched capacitor just act like a resistor again?

IMO it eliminates OP amp in signal path and most resistors which are most significant noise sources. It adds analog switches though...

[Kleinstein]

The current is too high to use is with normal integrators. Integrating ADC's use more like a current in the 100µA range, as at higher current the linearity of the OPs gets to bad. The direct integration of the current is more like a good option for very small current (pA range).

Just a 400 Ohms (or what ever fits the needed voltage range) resistor would give you a good conversion to voltage. If you can't tolerate the high voltage drop, the TIA is a good solution. The needed resolution is something like 3 µV - not easy, but not that difficult either. A current resolution in the 10 nA range is not that bad - even BJT based OPs could be OK. Leakage currents are more like in the 100 pA range and less. So no real need for guard lines - just a clean board. 20 Bit's resolution at kHz speed might be possible with good SD ADCs (LTC2440 or better) or other high end integrated ADCs. The LTC2440 is specified with a little over 3 µV RMS for a +-2.5 V range at about 2 ksps. So the ADC / amplifier part should get you where you want, even at 1 kHz.

[Alex Nikitin]

Unfortunately, there is no decent datasheet. The only other paper I have is calibration list. The only info about output is in attachment. R_L is between 100R and 130R.

From the look of it, you should be able to get away with just a resistive 400 Ohm load and a buffer for your ADC - mostly for protection of the ADC input and possibly to filter off some noise by limiting the bandwidth. No need for a TIA or any other complications. There is nothing more accurate than a good resistor to convert a current into a voltage  .

Cheers

Alex

[max_torque]

Surely the trick when operating below, or very close too the noise floor is to understand exactly what that noise looks like (in the time and frequency domains) so you can pull the real signal out from that noise?  Understanding the effect of various driving forces on the noise sources and calibrating those out via direct measurement of those driving forces seems like a reasonable approach to me?

[xorly]

From the look of it, you should be able to get away with just a resistive 400 Ohm load and a buffer for your ADC - mostly for protection of the ADC input and possibly to filter off some noise by limiting the bandwidth. No need for a TIA or any other complications. There is nothing more accurate than a good resistor to convert a current into a voltage  .

I was just unsure if resistor is ok. Current design (not mine) uses this method and there is significant noise on output. Probably not optimal PCB layout and/or component selection.

Then I need to get voltage output approx. +-2.5V. Just pass signal through buffer and divide?

By buffer, do you mean something else than voltage follower?

[Alex Nikitin]

I was just unsure if resistor is ok. Current design (not mine) uses this method and there is significant noise on output. Probably not optimal PCB layout and/or component selection.

Then I need to get voltage output approx. +-2.5V. Just pass signal through buffer and divide?

By buffer, do you mean something else than voltage follower?

Resistor is OK. The noise is most likely is originated in the accelerometer. The self-noise of 400 Ohm resistor in 400 Hz BW is about 130pA RMS which is low enough for your requirements. If you have access too the old design, check what the noise is with the accelerometer disconnected and the input shielded.

On the buffer - with 400 Ohm resistance it should be possible to connect the ADC with a simple RC filter, however the maximum output is over 30mA or 12V, which may damage the ADC, so you need some protection. A follower + LPF with 400Hz cut-off frequency would take care of it. There is no point in dividing the voltage, just use a suitable value resistor to get the full scale you need.

Cheers

Alex

[xorly]

I though of using voltage limits of current output as boundaries. Voltage divider with 1200 Ohm going from signal to 400 Ohm connected to GND.  Voltage sensing on 400 Ohm resistor. With 7.5mA, voltage on sensor output rises to 12V (approximately) and can't go higher.
But higher resistance, higher noise...

EDIT: math problems...

[Marco]

You might harm the linearity and/or bandwidth by doing it like that.

[xorly]

You might harm the linearity and/or bandwidth by doing it like that.

Can you explain source of these potential problems? Slew rate of current output? I can't see difference when we just change sensing resistor.

[Marco]

If there is some kind of active current source inside it might get more nonlinear the closer you get to it's compliance limits.

If it's passive the higher the load resistance the faster response should fall off I think.

[Cerebus]

If there is some kind of active current source inside it might get more nonlinear the closer you get to it's compliance limits.

If it's passive the higher the load resistance the faster response should fall off I think.

The 'datasheet' shows the bandwidth for the whole transducer to be 800 Hz. The datasheet leaves inferring whether this is a mechanical or electrical limitation (or both) as an exercise for the student.