Interfacing bipolar signal in single supply

[vvanders]

I'm in the process of building and adc frontend for one of Lattice's FPGA boards in order to mess around with some digital processing of audio. I'd consider myself okay at analog things but I'm having trouble differentiating what's important when it comes to dealing with the signal coming in to the ADC.

The reference board I have only supplies 3.3v and the input signal is +/- 0.9v with a possible impedance of 100-900 Ohms. I'm stuck at how I want to approach bringing this signal into my ADC in a range from 0V to Vref that will be accurate but also doesn't cross into the realm of audiophile phoolery.

Right now I'm looking at 3 possible options:

1. Use a summing configuration with an offset voltage shifting by 0.9v. This is the simplest option as I can take advantage of my ADC input op amp, however since I don't know the impedance of my source I could have some(potentially large) unknown offset. Buffering the input isn't an option since it would require a dual supply op amp to prevent clipping in the negative voltage.

2. Use a charge pump to generate a negative rail and then use standard op amp techniques to shift from negative range to 0V to Vref. I like this because it will be easy to buffer the input signal, however it means adding noise to the overall system and complexity that comes with a switching dc-dc voltage converter. Additionally it means extra op amps per channel, on the plus side I could also use this to have a dual output from my DAC as opposed to using a large capacitor in series and the roll-off that it incurs.

3. Creating a virtual ground on the input signal at 0.9V. This is the technique that I'm leaning towards the most, however I don't have quite a good enough grasp of how floating ground would work with an arbitrary input source. I've already prototyped it out in spice and everything behaves nicely:


The question I have is would the LT6220 have to provide enough output current to lift the ground for the entire circuit that's connected on the input or only for the ~600Ohm input that the circuit samples. If it's an unknown amount of current then this obviously won't work in the case I'm thinking of.

[kizzap]

what you are after is DC biasing.

have a look at this picture:

http://www.analog.com/library/analogDialogue/archives/35-02/avoiding/Fig1.gif (linked as I am not hosting it.)

the part you are interested in is the Cin, and Ra/Rb combo. The Ra/Rb combo puts a bias on the input voltage, which is effectively the same as your first solution

-kizzap

I should have linked in the page I found that image actually. has a bit of good information.

http://www.analog.com/library/analogdialogue/archives/35-02/avoiding/index.html

[T3sl4co1l]

What bandwidth?  Does the source expect to be terminated?  Does it have to be?  What ADC, or more precisely, what input specs -- SE, diff, voltage range, input impedance, etc.?  How many bits do you need?  What gain and offset accuracy is needed?  Can it be calibrated out?  (Should it be auto-calibrated?)  Also, will you be needing an antialias filter?

In absence of any details, I'd just make a voltage divider from +V (or VREF if available) to ADCIN to VIN.  A pair of 10k ought to do.  But do mind this introduces supply noise and halves the voltage swing, not to mention everything else.

Tim

[vvanders]

@kizzap: Yeah, that's essentially #1, however since the various inputs can have impedance from 100-600 Ohms that means that I could get unwanted offsets.

@T3sl4co1l: Audio range, although I'd like to try my hand at oversampling, so say up to 1msps, currently looking at LTC2365 but open to other options. The biggest issue seems to be that impedance of line out for most things isn't spec'd and can vary from 100-600Ohms. I don't expect what I build to be perfect but want to make sure that I'm at least considering all of the major factors.

[TMM]

@kizzap: Yeah, that's essentially #1, however since the various inputs can have impedance from 100-600 Ohms that means that I could get unwanted offsets.

It's AC coupled by the capacitor. The input impedance however is 50K rather than HighZ if that matters.

I've used a sink/source vreg (e.g. LT1118) to provide the virtual ground for an audio input in the same way as your schematic in the OP.

[T3sl4co1l]

Ah.  Then yes, AC coupled (presumably, a bandwidth of 20-20,000 Hz, so DC is not needed).  The pair of 10k resistors would do a fine job (audio outputs generally expect a load much higher than their rated output impedance, since an amp might have 10k or 100k input resistance for example).

An op-amp isn't required, but I'd put some protection diodes in there.  Never know what could be coming down the cable...

Tim

[vvanders]

Ah, I feel like an idiot missing the series capacitor. That's a much better way to deal with things.

@TMM: I thought about using a vreg after I posted the schematic, good to know that someone has used that approach in practice.

@T3sl4co1l: Datasheet calls for an ADC driver if impedance is high, or are you just talking about the level shifting part?

[T3sl4co1l]

If you still need a driver, then yeah, an op-amp would be good...

Not always the case; some act like a weak load (a few megs to ground).  The recommended impedance in that case (usually on the order of like 10k) is more to assure accuracy, especially while sampling many sources.  Others may 'destroy' the input voltage in the process and therefore act like a switched capacitor filter with beefier equivalent loading.

Tim

[vvanders]

Yeah, datasheet seems to call for one. I might play around with simulating the SAR ADC, seen some articles and might be a fun exercise.

AC coupling, did the trick, here's an updated schematic:


Did some AC analysis and low freq drop-off isn't too bad, I've used 220uF before for simple audio stuff, this will let me get on to the digital stuff which I'm sure will keep my busy for a while.

Thanks to everyone for all the help, analog isn't my strongest area for sure.