Active filter stage as driver for ADC good or bad?

My guess, off-the-cuff, would be that you don't need a voltage follower

- the active stage should provide plenty of drive for the ADC load.

But to be sure, compute the output impedance explicitly. Form a small-signal equivalent circuit model of the BP filter circult. You can do this using the ideal op-amp model, i.e., no current flows into the inputs and the positive and negative inputs are maintained at the same levels. Then short the (voltage) input and compute the output impedance by connecting a fictitious "test source" V_x at the output and determining the output current I_x that would flow into the output (back into the BP circuit). The output impedance is then V_x / I_x.

Armed with the knowledge of the output impedance, find the input impedance of the ADC on the datasheet. It should be at least a factor of

Electric,

I really depends on the output swing from your BP filter and the type of ADC that you are using. I would check the ADC's input specifications against your calculated BP filter out min and max to see if you need anything between them. It also depends on your specifications for regulation of the signal. If you need a very clean signal for high precision in the ADC then you may want to add another stage prior to the ADC. What type of signal will you be filtering and converting?

Keith

Thank you, I will try to do that. :-)

Just keep in mind that some ADCs chrage/discharge an on-chip cap that hangs right across the input. The driving stage must then be up to snuff to deal with that.

The ones that do that will tell you in the data sheets, and they'll tell you what level of charge injection you can expect when the ADC samples. What your filter/amplifier _really_ needs to do is to have that cap at the right voltage when the next sample commences.

The way that I've seen this done is to model the cap and the charge injection, then make sure that the filter/buffer settles to within the ADC's ENOB error within the ADC sample time -- plus a bit or two to accommodate the inevitable loss of precision when you go from mathemagic land to the real world.

that and what everybody else said. i would also think you wouldn't need to use a voltage follower between your BPF op-amp and the A/D as long as there are not passive components in between (the BPF has all of its compenents in the input and feedback of the op-amp).

the other thing i would say is keep in mind, in the signal processing that occurs after the A/D that it's on a filtered signal and not the flat one, in case you need to compensate for that somehow internally in the DSP. usually this is only the case when an anti-aliasing filter is a little too aggressive at the high end of the passband (below Nyquist) taht you wanna keep. perhaps in the OP's case he doesn't *want* to bring back anything that the BPF attenuated. i dunno.

r b-j

The most straight forward way to compensate for charge injection is to use an RC-filter as the last stage of the antialiasing filter, C being high enough to absorb the injected charge within < 1 LSB or so and tied directly to the ADC input. At higher speeds this might mean too low an R, but not in real life since the high-speed ADCs (those where this would be an issue) don't inject any charge into the input (at least those I have used and looked at).

Dimiter

------------------------------------------------------ Dimiter Popoff Transgalactic Instruments

------------------------------------------------------

Unless its a sigma-delta converter, or has explicit anti-alias filtering or buffer amps on board, pretty much any converter applies a god awful kick as the sampling gate opens. You not only need to ensure the output impedance of the filter is low enough for the sampling to settle to the required resolution within the gate time, you need to ensure those kicks don't destabilise the analogue circuitry you are adding. I've seen amps go kinda wacky when they are connected to an ADC.

People can be really woolly about the performance envelope of the things which drive ADC inputs, whether passive or active.

Regards, Steve

there is a bit in this article

Martin

Thank you all for your input :-)

Yep. It's one of my bread and butter jobs. My recipe is to never, ever, split ground planes. Ever. That plus a nice stiff low-Z drive has always yielded the performance the clients wanted. Before the design change they often had 2-3 bits worth or "weirdness" in the data. For those with ground splits the data looked like Niagara falls.

no probs. It's nice to try and help, but choose a better/personal nickname, if you are going to use newsgroups more often

Martin