I need a mini analog front end design review

[alexwhittemore]

Background: I'm a member of Boston University's Rocket Team. We make hybrid HDPE/N20 rocket motors. My task, electronically speaking, is to make a system capable of measuring the DC and AC components of a pressure transducer (well, many, and some load cells, but let's focus). The transducer is rated at 500PSI (sorry, it's aerospace here) FSO, and 3mV/Volt Excitation at FSO, where excitation is 10V. I need to measure the DC component accurately, as well as the AC component on top of that since the expectation is to glean the hemholtz frequency for the combustion chamber during the burn. I'm using an INA125 instrumentation amp to generate excitation for the transducer and amplify it with a gain of 100, such that a chamber pressure of 550PSI produces 3.3v, my ADC reference voltage. However, I need to low pass filter the signal at about fc=2kHz (but I'm not positive about that fc, so it needs to be tunable). My plan is to use the INA125 to drive a LT1063 switch-cap lowpass filter, which in turn drives my ADC, an Analog Devices AD7685 sampling at 20ksps. The ADC needs no additional input protection here, since the filter driving it has a maximum output current of 20mA (less than the 130mA the ADC can take if the input voltage is >.3v outside supply), however the filter will require something like a 20 ohm input resistor to prevent the INA125 from overdriving it in the case that the pressure transducer sees much more than 550psi (like in the case of a hard start). Finally of note, the LTC1063 runs on +5/0V supply and requires 4V on its clock pin for logic high, meaning my 3.3v microcontroller can't drive it: this will be rectified with a MC74VHC1GT126 TTL-input CMOS-output buffer from ON semi to buffer the clock line. The clock-to-corner ratio of the filter is 100:1, so for a 2kHz corner, the clock will need to be 200kHz.

Perhaps that will all be much clearer with an actual schematic: I should have one finished sometime tomorrow. In the mean time, the real question is, can this filter be operated effectively in this way, with a common mode on the input somewhere around 3 volts, preserving both that and the AC components on top of it? The datasheet doesn't make it totally clear that this will work, EG by providing a minimum output voltage rating for single-supply operation.

The fallback option is that I'm designing this whole analog front end on a small board to be placed as close to the sensors as possible. The plan is to put this filter assembly on a small daughtercard, such that if I mess up the filter design, I can replace only that component. As well, if the filter simply fails all over, I can run the amplified signal straight into the ADC with no filtering. This will at least allow logging of the DC pressure, and could give the particular frequency of interest so long as nothing in a similar range has aliased in.

[amspire]

In this type of circumstance, I might even consider a passive L/C filter. The passive filter will be able to cope with anything on the input, and will be immune to noise from the supply rails.  It is not tuneable, but is there a reason you need to limit yourself to 20KHz sampling?  Why not capture at 100KHz, grab more data, and use digital filtering on the results?  With a higher sample rate, then you could use a Bessel filter and still capture great quality data. 

If you need the most accurate waveform, then a constant delay filter like  Bessel filter may be best, but the stop band fall off is slow so.  If you just need to see the frequency components and the actual shape of the waveform is not important, then perhaps a Chebychev or Elliptical filter might do the job.

Butterworth filters like the LT1063 are like a compromise between Chebychev and Bessel filters. If the relative amount of noise near fc is relative small compared to lower frequencies, it make do an excellent job. If the noise near fc is relatively high, and you want an accurate waveform without spurious ringing in the waveform caused by the filter delays, then a Bessel filter may be the one you need.

I am just asking, as I really know nothing about the nature of the signal you are trying to capture.

Richard

[alexwhittemore]

Thanks for the reply! The shape of the waveform doesn't really matter, so phase linearity is not a requirement. I agree that oversampling would be nice (I could probably even get off with a one-pole RC) but since there are 4-5 of these running on the same SPI bus, I don't have the serial bandwidth.

Imagine you're blowing into a soda bottle - there's a ringing of air in steady state overpressuring, exiting, underpressuring, and reentering the bottle. This is true of any air crossing a barrier, so it's also true of the rocket motor. And it's dependent on volume of the cavity, which is what I need to calculate. But it's also got all the other audible noise of the rocket burning on top of that.

Bottom line, I'm looking for the ACCURATE amplitude DC value (overall chamber pressure) and a single frequency component that walks slowly downward in the 1-1.5 kHz range.

So the question is, is there anything to indicate that my setup may not work on a signal with DC around 3v and a small AC part on that?

[amspire]

Alex,

It may be just me, but I don't like the idea of overdriving the A/D inputs, even if it has built in protection diodes. I would limit befor the dvice, but lets leave that for the moment.

I have no problem in theory with that A/d being able to log data that you can average to get the DC, and you can run a FFT program (MATLAB or an open source alternative like Octave, Scilab or the MathCad like SMATH) to get the amplitude and frequency of the AC component.

However, you want accurate DC, and yet you are using an instrumentation amplifier with an offset of 250uV and so with the gain of 100, that makes 25mV. on top of that, the filter has another 5mv offset.

Since you frequency requirements are low, I would consider looking at Auto-Zero or chopper stabilized amps - perhaps like the AD8553 from Analog Devices - or you can devices with offsets near 1uV if you need it.  If you can get a single +5v amp and it feeds the A/d running at 5V, then you never overload the inputs.

Now for the filter, I would still favour a passive filter between the amp and the a/d. You have plenty of space between a 2kHz fc and the sampling rate of the A/D, and a passive filter has zero offset. You may even be able to use a simple RC filter. I would add  a simple RC filter on the inputs of the amp as well. As logs as the 1.5KHz Ac ends up being significantly larger in amplitude at the A/d to any stopband noise, then the FFT will correctly see the Ac signal. I would dump the filter IC. Putting a dual supply filter chip between two single supply chips is not a perfect solution.

It really comes down to the nature of the noise in the stop band. Is it very high amplitude , or is it of the same order of magnitude to the AC 1 to 1.5KHz signal?

Richard

[alexwhittemore]

The protection diodes are spec'd at 130mA overdrive handling, so a 20 ohm resistor in series with the input should only minimally affect frequency response but will limit current to 90mA in the worst case overdrive that the system will produce without physically exploding (hookup errors are a different deal, but the kind of overdrive I'd have to handle to protect against them is much worse, so I'm not going to).

[amspire]

Yes, I know the A/d has protection diodes, and yes, they are rated at 130mA.

But AD says something like don't exceed the rail by more than 0.3V.  If you do, there are protection diodes.  When I read that, I read that the protection diodes make sure that the device doesn't get destroyed, but there is no certainty that the A/D will be function properly if overdriven. It is not designed to be overdriven in normal use.

It is just one of those little decisions that can bite you unexpectedly later on - "Who would have guessed the A/D would do that!".  Weird things can happen like the protection diode causing local heating that causes an error in the A/D linearity.  As far as AD is concerned, in a good design, the diodes are never used. All the data sheet says is that if you put 130mA through the diode, the A/D won't be destroyed.

I just find that it is a good policy to use devices like A/Ds exactly as the manufacturer intend them to be used.

And, as I suggested, it is possible that a passive filter can do a better job in this situation then the active filter, but then again, maybe not. If you need an active filter, why not just get a dual single supply Auto zero opamp package and use it for a 5th order sallen-key Bessel filter?  Best of all worlds. No negative voltages. Low offset.  Best waveform integrity, and probably much cheaper.

Richard

Richard.

[alexwhittemore]

Re:diodes- true all. But A. the input protection resistor is specified such that that 130mA won't be exceeded in the WORST case - the system will literally explode, likely hurling shrapnel everywhere, if the A/D gets enough input to put it over. Point being, during normal operation, input won't exceed vref=3.3v. And for that matter, input will only be out of spec if things in the motor get SERIOUSLY out of spec. So I'm hardly worried about collecting data up there. It'd just be nice if there wasn't permanent damage is all.

Re:filter- Also true. And I'm sure you're right. But you're designing the good solution and the volume solution. I'm only buying a couple of these parts, and our budget is huge compared to electronics cost. In fact, we would (and have previously) just bought premade daq solutions from the likes of NI. The problem there is two fold: one, that doesn't take care of control, which stil relies on some other system to handle switching valves and relays, and two we hope to eventully (one or two years out) put the motor in a body and launch it in the desert. We still need to be collecting data then. Hence size is an issue (albeit only insofar as we can't fit NI boxes and a computer in the thing).

Development time is the real key at this point. I want to get the thing out the door. And while filtering is the proper thing to do, I've actually just designed in a daughter card system so filter circuits can be swapped or bypassed. The black box solution is easiest, and at this point that's what I need.

But while we're talking about easy, I'd love to hear suggestions for black box or prebuilt fixed filter solutions. I've seen some other active op-amp based parts, and I know fully passive parts exist, but I haven't found any that look as easy. I'm totally happy to spend time on making a second filter design to swap in when this one hits the fan, if you will.

[ejeffrey]

I would switch instrumentation amplifiers to something with better DC precision, and replace the filter with something like the LT1069.  That gives you an 8th order elliptic low-pass filter.  You don't have to make it tunable, just set the cutoff frequency at around 75% of the nyquist frequency.  That will stop aliasing, and your analysis software will pick the 2 kHz tone out with no problem.  I would still add a passive RC filter at the input to the instrumentation amplifier, just to avoid amplifying more high frequency noise than necessary.  How good do you need your DC accuracy?  A couple of microvolts is what you are likely to get from a well chosen instrumentation amplifier. That would give you a .1% offset error on your 3 mV full scale.  A chopper stabilized amp like amspire recommends will virtually eliminate DC offset, but keep in mind that the clock frequency may end up in the same range as the resonance you are trying to detect.  Clock feedthrough should be small, but I don't know how big your signal is.  I assume you have a differential signal from something that looks like a strain gauge, so you need a differential amplifier with high common mode rejection.  That means you would need a chopper stabilized instrumentation amplifier, not just an op-amp.