Anti-aliasing filter for Ultrasound application

[nickagian]

Hi all,

I am designing an application that measures the outputs from ultrasound transducers. The data are to be processed at a low-power MSP430 microcontroller. Before the MCU, a high-speed 12-bit ADC will be used. It's highest possible sampling rate is 1MSPS. I would need some help with the design of the anti-aliasing filter.

I am aware that the optimum solution for the filter characteristics would be such that it's response is down more than 20log(1/4096)dB = -72dB at the Nyquist frequency (500kHz). However, this would reduce the effective bandwidth of the signal from the sensors to less than 200kHz and I'm not yet sure if this acceptable from the application's level. But anyway, if we assume that the attenuation of the filter is not around -72dB but worse, what should be an acceptable level? For example, if the attenuation would be around -40dB, would that be acceptable? Or -put in another way- how much error would be introduced at the digitization if at the Nyquist frequency the loss is only -40dB? How can I calculate that?

Finally, what filter type is the best solution for an anti-aliasing filter? The most standard solution would be the Butterworth, but what about the Chebysev or the Elliptic? I could achieve better attenuation beyond the passband with these filters. But would the ripple at the passband and the larger phase group delay?

I appreciate any help that you can give me with these issues.

Nick

 

[crutschow]

The answer to these questions really depends upon your application. For that, detailed information about your signal and noise requirements is needed. You have not given that information.

The required attenuation of an anti-alias filter is determined by the amount of undesired signal/noise outside the passband you must suppress. 72dB of attenuation is needed if the undesired signal/noise is equal to the signal. If it is just random noise, then you could use less attenuation. You just need enough attenuation to keep it below 1 LSB at the input to the A/D. Thus you need knowledge and/or measurement of the undesired signal/noise level to determine the required attenuation.

Chebysev or the Elliptic filters do give better attenuation outside the passband (as well as would making the filters higher order than 2-pole). Whether the ripple or phase group delay will be a problem is determined by the bandwidth of the desired signal and how accurately you need to measure the signal amplitude and phase over this bandwidth.

Only you can answer those questions.