Aliasing is an effect which occurs when sampling an analog signal. It is important to recognize that there is no way to prevent this effect from happening. The only thing we can do is to minimize its effect.
A common misunderstanding is that aliasing can be prevented using a “high enough” sampling rate, making sure that the frequencies of interest lay below the Nyquist Frequency. For acoustics one might, for example, argue that 50-100 kHz should be more than sufficient to capture all relevant (NVH) frequency content. There is however no guarantee that sensors / cabling won’t pick up (EMI) noise with operating frequencies in Mega or Giga Hertz region. Picked-up, these very high frequencies Mirror / Alias onto your FFT spectrum falsifying your analysis.
So what should front-ends do to minimize the effect of aliasing? The answer lays not in digital signal processing but in the analog world: Application of a proper analog low-pass filter. Indeed, once transformed into the digital domain, there is nothing one can do anymore to distinguish aliasing effects. As a consequence we can only tackle the problem in the analog world.
A low-pass filter (in this context commonly referred to as anti-aliasing filter) can be a relative simple circuit of resistors and capacitors. The circuit makes sure that frequencies above the maximum frequency of interest (Nyquist Frequency) are severely reduced in amplitude (reduction in the order of a million for example). Although therefor still present after sampling, the effect on the FFT spectrum is minimal.
A typical anti-aliasing (i.e. low-pass) filter is shown in the figure above. It has a flat response below the Nyquist Frequency with a gain equal to 1. Above the Nyquist frequency, the filter kicks in and tries to diminish any frequency content.
Please notice that there will always be a certain “grey area” approaching the Nyquist Frequency where the signal is not yet fully attenuated. The low-pass filter simply can’t create a complete sudden drop at the Nyquist Frequency in practice. For this reason, very often dynamic analysers use a factor less than the Nyquist Frequency (½ sampling frequency). In PAK, a “safety factor” of the sampling frequency divided by 2.56 is applied making sure that our module specifications are met in the entire bandwidth.
In the past, engineers designed (analog) anti-aliasing filters of high order, e.g. consisting of many electric components, to create a steep descent above the Nyquist Frequency. This made these devices extremely expensive, temperature sensitive and big.
In the next edition, multi-rate sampling techniques applied in modern analyzers will be discussed. Please feel free to suggests further topics for upcoming PAK InfoLetters.