Forums

Anti-Aliasing filters with sigma-delta A/Ds

Started by HardySpicer April 24, 2012
Do you still require an anti-aliasing filter with sigma-delta A/D's? I
am guessing yes but very low order.
A/D I am testing seems to have an excellent roll-off without any
filter at all.


Hardy
On 24.04.2012 8:38, HardySpicer wrote:
> Do you still require an anti-aliasing filter with sigma-delta A/D's? I > am guessing yes but very low order. > A/D I am testing seems to have an excellent roll-off without any > filter at all. > > > Hardy
A simple RC filter would suffice in most cases. You may find some useful information in this article: http://www.analog.com/library/analogDialogue/Anniversary/15.html -- Alexander
>On 24.04.2012 8:38, HardySpicer wrote: >> Do you still require an anti-aliasing filter with sigma-delta A/D's? I >> am guessing yes but very low order. >> A/D I am testing seems to have an excellent roll-off without any >> filter at all. >> >> >> Hardy > >A simple RC filter would suffice in most cases. You may find some useful >information in this article: >http://www.analog.com/library/analogDialogue/Anniversary/15.html > >-- > >Alexander
That's a good page, who's critically important part is http://www.analog.com/library/analogDialogue/Anniversary/Graphics/figure59lg.gif . You want a capacitor between the inputs of a size that gives appropriate rolloff in combination with the resistors. This filter has almost no ability to inject differential noise, so its very tolerant of board layout. Then you want two very small capacitors from the 2 inputs to ground, to bypass RF. Without those you may have problems when you try irradiating the board with a few hundred MHz. The main filter just can't cope with that because a) the capacitor across the inputs is probably of the kind of value that sucks at RF, and b) the differential quality of the chip's inputs falls apart at sufficiently high frequencies. Steve
On Mon, 23 Apr 2012 21:38:36 -0700, HardySpicer wrote:

> Do you still require an anti-aliasing filter with sigma-delta A/D's? I > am guessing yes but very low order. > A/D I am testing seems to have an excellent roll-off without any filter > at all.
You need an anti-alias filter, but it only needs to hold the aliases down at the sigma-delta sampling rate (which is fast), and have a rolloff that is acceptable at the speed at which you sample out of the ADC (which is slow). So generally an RC is enough, although of course you should still think about it. -- Tim Wescott Control system and signal processing consulting www.wescottdesign.com
On 4/24/12 9:40 AM, steveu wrote:
>> On 24.04.2012 8:38, HardySpicer wrote: >>> Do you still require an anti-aliasing filter with sigma-delta A/D's? I >>> am guessing yes but very low order. >>> A/D I am testing seems to have an excellent roll-off without any >>> filter at all. >>> >>> >>> Hardy >> >> A simple RC filter would suffice in most cases. You may find some useful >> information in this article: >> http://www.analog.com/library/analogDialogue/Anniversary/15.html >> >> -- >> >> Alexander > > That's a good page, who's critically important part is > http://www.analog.com/library/analogDialogue/Anniversary/Graphics/figure59lg.gif > . You want a capacitor between the inputs of a size that gives appropriate > rolloff in combination with the resistors. This filter has almost no > ability to inject differential noise, so its very tolerant of board layout. > Then you want two very small capacitors from the 2 inputs to ground, to > bypass RF. Without those you may have problems when you try irradiating the > board with a few hundred MHz. The main filter just can't cope with that > because a) the capacitor across the inputs is probably of the kind of value > that sucks at RF, and b) the differential quality of the chip's inputs > falls apart at sufficiently high frequencies.
i'm curious what the RC time constant comes out to be. if the two R's are the same and the three C's are the same, is the corner frequency equal to 1/(2*pi*(2R)*(3/2 C)) ? the two R's are in series and the C across the diff input is in parallel to the two other C's in series. is that how it works out? the one thing that worries me about a diff input driven by two separate op-amp buffers is a mismatch of gain between the two buffers. if they're mismatched, you won't have much CMRR. maybe if the op-amps were true envelope followers (rather than inverting amps as depicted) there would be little of that. -- r b-j rbj@audioimagination.com "Imagination is more important than knowledge."
On Apr 25, 1:48&#2013266080;am, Tim Wescott <t...@seemywebsite.please> wrote:
> On Mon, 23 Apr 2012 21:38:36 -0700, HardySpicer wrote: > > Do you still require an anti-aliasing filter with sigma-delta A/D's? I > > am guessing yes but very low order. > > A/D I am testing seems to have an excellent roll-off without any filter > > at all. > > You need an anti-alias filter, but it only needs to hold the aliases down > at the sigma-delta sampling rate (which is fast), and have a rolloff that > is acceptable at the speed at which you sample out of the ADC (which is > slow). > > So generally an RC is enough, although of course you should still think > about it. > > -- > Tim Wescott > Control system and signal processing consultingwww.wescottdesign.com
Can you not therefore rely on the "natural" roll-off due to amplifiers etc if you are dealing with "low" frequencies. eg my frequencies are only up to 20kHz and my clock is 40MHz. Hardy
On Tue, 24 Apr 2012 14:37:17 -0700, HardySpicer wrote:

> On Apr 25, 1:48&nbsp;am, Tim Wescott <t...@seemywebsite.please> wrote: >> On Mon, 23 Apr 2012 21:38:36 -0700, HardySpicer wrote: >> > Do you still require an anti-aliasing filter with sigma-delta A/D's? >> > I am guessing yes but very low order. >> > A/D I am testing seems to have an excellent roll-off without any >> > filter at all. >> >> You need an anti-alias filter, but it only needs to hold the aliases >> down at the sigma-delta sampling rate (which is fast), and have a >> rolloff that is acceptable at the speed at which you sample out of the >> ADC (which is slow). >> >> So generally an RC is enough, although of course you should still think >> about it. >> >> -- >> Tim Wescott >> Control system and signal processing consultingwww.wescottdesign.com > > Can you not therefore rely on the "natural" roll-off due to amplifiers > etc if you are dealing with "low" frequencies. eg my frequencies are > only up to 20kHz > and my clock is 40MHz.
Very possibly. Make sure you don't have any late-stage high frequency noise being introduced, and if not -- go to town. -- My liberal friends think I'm a conservative kook. My conservative friends think I'm a liberal kook. Why am I not happy that they have found common ground? Tim Wescott, Communications, Control, Circuits & Software http://www.wescottdesign.com
On Apr 25, 10:10&#2013266080;am, Tim Wescott <t...@seemywebsite.com> wrote:
> On Tue, 24 Apr 2012 14:37:17 -0700, HardySpicer wrote: > > On Apr 25, 1:48&#2013266080;am, Tim Wescott <t...@seemywebsite.please> wrote: > >> On Mon, 23 Apr 2012 21:38:36 -0700, HardySpicer wrote: > >> > Do you still require an anti-aliasing filter with sigma-delta A/D's? > >> > I am guessing yes but very low order. > >> > A/D I am testing seems to have an excellent roll-off without any > >> > filter at all. > > >> You need an anti-alias filter, but it only needs to hold the aliases > >> down at the sigma-delta sampling rate (which is fast), and have a > >> rolloff that is acceptable at the speed at which you sample out of the > >> ADC (which is slow). > > >> So generally an RC is enough, although of course you should still think > >> about it. > > >> -- > >> Tim Wescott > >> Control system and signal processing consultingwww.wescottdesign.com > > > Can you not therefore rely on the "natural" roll-off due to amplifiers > > etc if you are dealing with "low" frequencies. eg my frequencies are > > only up to 20kHz > > and my clock is 40MHz. > > Very possibly. &#2013266080;Make sure you don't have any late-stage high frequency > noise being introduced, and if not -- go to town. > > -- > My liberal friends think I'm a conservative kook. > My conservative friends think I'm a liberal kook. > Why am I not happy that they have found common ground? > > Tim Wescott, Communications, Control, Circuits & Softwarehttp://www.wescottdesign.com
so convenient these Sigma-Delta A/Ds. Best things since sliced FPGAs! Hardy
On Tue, 24 Apr 2012 11:04:04 -0400, robert bristow-johnson wrote:

> On 4/24/12 9:40 AM, steveu wrote: >>> On 24.04.2012 8:38, HardySpicer wrote: >>>> Do you still require an anti-aliasing filter with sigma-delta A/D's? >>>> I am guessing yes but very low order. >>>> A/D I am testing seems to have an excellent roll-off without any >>>> filter at all. >>>> >>>> >>>> Hardy >>> >>> A simple RC filter would suffice in most cases. You may find some >>> useful information in this article: >>> http://www.analog.com/library/analogDialogue/Anniversary/15.html >>> >>> -- >>> >>> Alexander >> >> That's a good page, who's critically important part is >> http://www.analog.com/library/analogDialogue/Anniversary/Graphics/
figure59lg.gif
>> . You want a capacitor between the inputs of a size that gives >> appropriate rolloff in combination with the resistors. This filter has >> almost no ability to inject differential noise, so its very tolerant of >> board layout. Then you want two very small capacitors from the 2 inputs >> to ground, to bypass RF. Without those you may have problems when you >> try irradiating the board with a few hundred MHz. The main filter just >> can't cope with that because a) the capacitor across the inputs is >> probably of the kind of value that sucks at RF, and b) the differential >> quality of the chip's inputs falls apart at sufficiently high >> frequencies. > > i'm curious what the RC time constant comes out to be. if the two R's > are the same and the three C's are the same, is the corner frequency > equal to > > 1/(2*pi*(2R)*(3/2 C)) ? > > the two R's are in series and the C across the diff input is in parallel > to the two other C's in series. is that how it works out?
Yes, for differential mode signals (i.e. the signals you want). For common mode signals, the time constant = RC
> the one thing that worries me about a diff input driven by two separate > op-amp buffers is a mismatch of gain between the two buffers. if > they're mismatched, you won't have much CMRR. maybe if the op-amps were > true envelope followers (rather than inverting amps as depicted) there > would be little of that.
1% resistors are very cheap (< 1 cent), whereas capacitors don't usually have a tolerance of better than 10% or so. 1% is possible, but expensive. Mismatch between the two caps to ground will result in common mode to differential conversion that is frequency dependent. This may not matter if you are only interested in a restricted band, e.g. < 20kHz, but watch out for your switching power supply (50-500 kHz) ending up in-band after sampling. Regards, Allan
>On 4/24/12 9:40 AM, steveu wrote: >>> On 24.04.2012 8:38, HardySpicer wrote: >>>> Do you still require an anti-aliasing filter with sigma-delta A/D's?
I
>>>> am guessing yes but very low order. >>>> A/D I am testing seems to have an excellent roll-off without any >>>> filter at all. >>>> >>>> >>>> Hardy >>> >>> A simple RC filter would suffice in most cases. You may find some
useful
>>> information in this article: >>> http://www.analog.com/library/analogDialogue/Anniversary/15.html >>> >>> -- >>> >>> Alexander >> >> That's a good page, who's critically important part is >>
http://www.analog.com/library/analogDialogue/Anniversary/Graphics/figure59lg.gif
>> . You want a capacitor between the inputs of a size that gives
appropriate
>> rolloff in combination with the resistors. This filter has almost no >> ability to inject differential noise, so its very tolerant of board
layout.
>> Then you want two very small capacitors from the 2 inputs to ground, to >> bypass RF. Without those you may have problems when you try irradiating
the
>> board with a few hundred MHz. The main filter just can't cope with that >> because a) the capacitor across the inputs is probably of the kind of
value
>> that sucks at RF, and b) the differential quality of the chip's inputs >> falls apart at sufficiently high frequencies. > >i'm curious what the RC time constant comes out to be. if the two R's >are the same and the three C's are the same, is the corner frequency >equal to > > 1/(2*pi*(2R)*(3/2 C)) ? > >the two R's are in series and the C across the diff input is in parallel >to the two other C's in series. is that how it works out?
Why would you make the three Cs the same? To get lousy results? I think you should stick with what I wrote. :-)
>the one thing that worries me about a diff input driven by two separate >op-amp buffers is a mismatch of gain between the two buffers. if >they're mismatched, you won't have much CMRR. maybe if the op-amps were >true envelope followers (rather than inverting amps as depicted) there >would be little of that.
Steve