Post Go back to editing

lower ADXRS620 output impedance , higher ADC accuracy ?


I use LTC1408 (ADC) to convert ADXL325 and ADXRS620 analog signals. LTC1408 ask to use a buffer amplifier before ADC to transform the analog signals, which have high source impedance.

ADXL325 has an output impedance of 32k for LP filter, so it is necessary to add a buffer to the signal.

For ADXRS620, the temperature output impedance is very high, so it is also needed to put a unit buffer to the signal before ADC.

My question is where to know the impedance of the gyro's rate signal. Is it high or low? Is it essential to use a unit buffer to the signal if a high ADC accuracy is required?

Thanks in advance

  • The ADXRS620 rate output impedance is quite low. Under ~100 Ohms. It's unlikely you'll need to buffer it ahead of an ADC, but be careful with your anti-aliasing filter.

  • Hi, I appreciate your previous reply.

    I believe that nearly all applications have to use anti-aliasing filter for MEMS Acc and Gyro. If I use RC lowpass filter, the DC output impedance will increase significantly and AC output impedance should be small.

    For high sampling rate ADC, like 10kHz, I would like to verify if the active lowpass filter is the only solution to guarantee ADC accuracy. Am I correct to assume that using an active lowpass is the only way to maintain low output impedance while ensuring anti-aliasing filtering in a vibration-dynamic environment?

  • Hi Guillaume - glad that my response was helpful. Your last question though, is a lot more complicated!

    Let's start by looking at the function of an anti-aliasing filter. You want to keep unwanted high frequency content out of the ADC. There are many reasons why you would want to do that, the most common ones are:

    1) You might not be interested in information past some frequency.

    2) You might have some spurious high frequency content in the signal (like noise, for example) that you want to keep out of the ADC.

    Let's look at (1). In a mechanical system (your gyro is connected to a mechanical system, isn't it?) bandwidth is normally quite low as the mass is relatively high. Anything past some hundreds of Hz is most likely of no interest. Of course, you could just have the ADC sample much faster (say 100kHz) and add a digital filter to remove unwanted high frequency content. So no additional anti-aliasing filtering is necessary. This is usually my prefered route.

    In the case of (2) You would want to filter out signals of higher frequency than half the sample rate of the ADC to avoid having the high frequency signals folding back into the baseband (where they would be impossible to sperate from real information). But where is this high frequency content coming from? As a practical matter noise in our inertial sensors is almost completely random. So what happens if a sample here or there is "incorrect" due to aliasing? Probably nothing as over any reasonably long time this error integrates out to zero. Assuming you use the gyro correctly (have a clean power supply, for example) there really aren't any ways of getting periodic high frequency signals out of it. Also, practically speaking, the bandwidth of our gyros is a few kHz at best and most ADCs sample at more than 10 times this rate. So high frequency content coming out of the gyro is not much a problem.

    Of course, this is a simplified view of anti-aliasing. A careful designer would want to take into consideration many other factors. If you are looking for very high resolution, for example, you need to ensure that your input impedance into the ADC is very low to allow the ADC input stage to settle fast enough. So passive anti-aliasing filters are probably out of the question. On the other hand if all you are looking for is a 12 bit converter a passive AAF is probably just fine - particularly if the system bandwidth is low. Like I said before, it's a complicated question that is not easily answered by generalities.

  • Hi Harvey,

    I am very grateful for your insightful and clear explanation on the topic.

    I use MEMS gyro and acc as well as 14 bit ADC for flight state-estimation, under high in-flight vibration conditions. The targeted dynamics range is below 30Hz. To avoid aliasing, I think that it is ideal to have high sampling rate and FIR filtering. I am going to build two sampling systems, one using 400Hz and the other10kHz, to verify the INS performance. I think that I should use the active AAF for the 10kHz sampling system, but is it essential to use active AAF for the 400Hz sampling system? I hope to hear your opinions on this.

    Thanks in advance.

  • Hi Guillaume,

    First off, I assume you'll set the gyro bandwidth to about 40Hz or below.

    If you have a 10kHz ADC you are oversampling by a factor of about 250. Even with no AAF and a very simple digital filter you'll have >5kHz energy attenuated by more than 40db. Sounds like a lot, but that's only equivalent to about 7 bits. The good news is that you can do a lot better with a more sophisticated digital filter and get >70db of attenuation, which is about as good as you'll need. I would tend towards being conservative and just put a simple RC AAF consisting of a 4.7µF cap and a 600 Ohm resistor between the gyro and ADC as well as use a reasonable digital filter. The 600 Ohm series resistor is small enough that it should not wreck the settling time of the ADC input stage.

    In the case of the 400Hz ADC you are only oversampling by a factor of 10 so you can't make as steep a digital filter. Here the AAF appears more important. So I would add a few poles of active filtering. (Once you get past two poles passive filters start becoming problematic, so you'll probably want to go with an active filter.)

    The above assumes that you may have some periodic mechanical noise (resonances) in your system, some of which may be high Q. In fact maybe you do, maybe you don't. Since you will actually prototype the two systems and compare performance you have the luxury of not having to make the right decision based on incomplete information. But if I had to guess what to do without good information about the mechanical system I would go with the 10kHz sampling system.

    Best of luck!