Anti aliasing

[dpj]

Want to understand effect of ANTI ALIASING, while sampling for Voltage and current.
I am designing a e meter using MSP430 chip.
And want to lay Analog sensing part can anyone help me to understand how exactly this ANTI ALIALISING works?
Thanks and regards.

 

[Nigel Goodwin]

A steep lo-pass filter at less than half the sampling rate.

However, for simply monitoring current and voltage in a PSU it's unlikely to be needed - basically it's for if you're trying to record audio etc.

 

[ronsimpson]

If the power line was a good 50/60hz with out harmonics then no filter is needed. While the voltage is mostly sign wave the current is likely very much not sign wave. Some meters sample at 2khz to 20khz to measure the current more accurately. The current has high frequency noise that will upset the ADC if allowed in.

You should search for a example of "ANTI ALIASING".

 

[crutschow]

The problem is that any signal or noise about 1/2 the sample frequency will be captured and then appear as a lower (aliased) frequency in the sampled which then can not then be filtered out by any digital filtering of the sampled data. Thus you need a filter that minimizes any frequencies above 1/2 the sample frequency to avoid the aliased noise. To minimize the steepness requirement of the alias filter rolloff you want to sample the data significantly faster than twice the highest frequency of interest. That way the filter rolloff can be set well below the alias point of 1/2 the sample frequency.

 

[Mr RB]

Aliasing can also mean that samples taken on a slowly changing slope can be rounded up/down to the nearest ADC resolution bit, if that is what the OP is asking about then I'd leave some HF noise in place (don;t use too much low pass filtering) and sample at a higher freq than the desired samplerate, then average samples in software to give you higher resolution ADC data.

For instance if you need 1000 samples a second then sample at 8000 samples per second, and add each set of 8 samples together. It will do a similar job to hardware filtering but will remove the aliasing caused by the ADC rounding.

 

[crutschow]

Aliasing can also mean that samples taken on a slowly changing slope can be rounded up/down to the nearest ADC resolution bit, if that is what the OP is asking about then I'd leave some HF noise in place (don;t use too much low pass filtering) and sample at a higher freq than the desired samplerate, then average samples in software to give you higher resolution ADC data.

For instance if you need 1000 samples a second then sample at 8000 samples per second, and add each set of 8 samples together. It will do a similar job to hardware filtering but will remove the aliasing caused by the ADC rounding.

I've never heard of aliasing being related to ADC quantizing or rounding error, only to undersampling, which generates aliased signals. Do you have a reference for that?

 

[Mr RB]

Hi Crutschow, aliasing also occurs when you sample an analog line or curve in pixel or ADC sample units, they cannot follow the analog shape and default to the nearest digital unit. This produces an aliasing effect that basically generates another "alias" or fake frequency made from the step units. It's a common problem in graphics and audio sampling in the software world;

**broken link removed**
**broken link removed**

I'm not sure that's what the OP was asking? But it is inline with trying to get good AC sample data on an ADC system.

 

[misterT]

Aliasing can also mean that samples taken on a slowly changing slope can be rounded up/down to the nearest ADC resolution bit, if that is what the OP is asking about then I'd leave some HF noise in place (don;t use too much low pass filtering) and sample at a higher freq than the desired samplerate, then average samples in software to give you higher resolution ADC data.

Isn't that called dithering rather than anti-aliasing?

To OP:
Here is a great article about Anti-Aliasing, Analog Filters for Data Acquisition Systems.

 

[crutschow]

Yes, I consider picture artifacts, such as moire patterns, to be due to aliasing but I hadn't considered the stairstep pattern in a diagonal line as an aliased effect. However apparently that term is used, in general, to describe that also.

 

[Mr RB]

Yeah it's probably more accurately described as quantising noise or error, but in the software world I think they just use the term "aliasing" to describe any situation where sampling ADC etc makes an imperfect result.

(re oversampling and averaging) ... Isn't that called dithering rather than anti-aliasing?

I think we're all guessing what the OP meant when he said "anti aliasing", but having done an energy meter design I remembered the issue of the small voltage from the current sensor giving a very "steppy" ADC waveform and using oversampling etc to try and reduce the problem. So I guessed that might be what the OP meant and took a gamble, it was something that had not really been covered in the thread.

 

[dougy83]

In both of the above references aliasing referred to aliasing in images. As mentioned by MisterT, utilising the HF noise would be called dithering; the averaging of the oversampled signal provides the "noise-shaping" (LPF of the HF noise). A "steppy" waveform is the result of quantisation error.