amplifier and filter

[Gram]

I need basically to measure the acceleration response of a steel rod subject to a short duration impulse generated by metal-to-metal impact to the tip of the rod. The natural frequency of the steel rod is about 10 kHz. I have an ISOTRON Accelerometer-Model 7250A-10 from ENDEVCO and a Charge Conditioning Amplifiers Type 2635 from B&K. It would be very grateful if anybody can advise me whether I can use 2635 as a current power source and analog filter to this ENDEVCO accelerometer, which is with built-in electronics.

[Gram]

As a civil engineer, I am short of knowledge in this area. Could anybody also please kindly answer the following questions for me?

1. If an accelerometer has a frequency response (as stated on its spec), say, 2-20,000 Hz, however I am going to acquire data at 100 kHz. If the output data contain some frequency components exceeding 20 kHz, are the responses beyond 20 kHz correct? Must I get rid of the higher frequency components through filtering?

 

2. If a filter is really necessary, can I just apply a digital filter by LabVIEW filter vi, to remove the vibration data above 20 kHz. Or must I have an analog filter on the front end of the data acquisition system.

3. In case an input excites transducer resonance, what would be the response of an accelerometer, or can I still use the result?

[Gram]

the accelerometer and charge amplifier can be found at

http://www.bksv.com/pdf/7250A-2-10.pdf#search='ISOTRON%20AccelerometerModel%207250A10%20from%20ENDEV...

http://www.testequip.com/sale/used/catalogs/B&K/0/B&K_2635.pdf#search='Charge%20Conditioning%20Ampli...

thanks a lot in advance.

[Buechsenschuetz]

1. There will be a certain error when measuring signals with high frequencies and an insufficient sampling rate. The error will encrease the closer the signal frequency comes to the sampling rate. You can also calculate this error (for sinusoidal signals) - I don't have the equation at hand. Nyquist's theorem says that you need twice the signal's frequency for sampling but this only applies to sinusoidal signals.

Also, signals with frequencies close or above the sampling rate will cause an effect called aliasing. Similar to an optical moire effect the 'mix' of signal and sampling frequencies will generate fake signals.

It is a general rule in data acquisition to limit the bandwidth of the incoming signal as far as possible, i.e. as far as your signals allow.

2. Of course a digital filter will just filter the results of the ADC. Aliasing signals generated by the ADC cannot be filtered out this way. It is good practice to use a hardware filter on the input. It will also solve problems like overdriving the inputs with signals or spikes with frequencies outside the needed range.

[Gram]

Many thanks for your reply. Specially, I have a DAQ which has a high bandwidth, 1 MHz. The bottleneck is the transducer with a frequency response between 2 Hz-20 KHz. Can I sample the transducer at 100 KHz without an analog filter, as I do not have a analog filtering device at hand?

[Buechsenschuetz]

It is possible to so, but it is not advisable. You cannot limit the bandwidth of the ADC by reducing the sampling rate. This will only reduce the bandwidth which is sampled without error and aliasing. Any signals outside the frequency range will be sampled, too, but the result will be a 'fake' signal which looks as if it were inside the bandwidth of the original signal.

I found a nice demonstration of aliasing:

http://www.dsptutor.freeuk.com/aliasing/AD102.html

If you enter an input frequency of 20000Hz, the ADC will keep sampling with its rather low rate (check the 'grid' box to show the sampling intervals). At the sampling intervals the ADC will measure the actual voltage of the input signal - but this will be different at each sampling point. Assuming that both the input signal and sampling frequency are stable, the result will be sampled values which show a sinusoidal wave of much lower frequency. Of course it is not possible to tell whether this is a result of aliasing or an actual input signal.

So probably any kind of input filter will be better than nothing. You can build a very simple first order filter with a series resistor and a capacitor across the input pins.

[Buechsenschuetz]

PS. You should check both the 'grid' and the 'sampling points' boxes in the demonstration. To start the demo, check the 'plot' button.

[Buechsenschuetz]

If you do not have a hardware filter on the input, I think it is better to use a high sampling rate and do digital filtering. A high sampling rate will avoid aliasing as long as the signal frequency is less than, say, 1/2 the sampling frequency.

[Gram]

Thanks kindly, Buechsenschuetz, The low pass filter can effectively eliminate aliasing singles. If digital filter cannot replace analog filter, what is the point for using a digital filter? Just decimating and hence reducing the data size? or make the curve smooth?

[Logan_K]

I looked at the datasheet for your charge amplifier, and it has an adjustable lowpass filter on its output which appears to have 30kHz and 100kHz cutoff frequency settings. If you set the 30kHz setting, then you will pass the full bandwidth of the accelerometer through, and as long as you sample at 60kS/sec or faster you will avoid aliasing. So I think you will be ok.

Regards,

Logan Kunitz