Multiplying DAC frequency response plot question

[CopperCone]

I am a bit confused by this plot in the LTC1595

http://cds.linear.com/docs/en/datasheet/159561fc.pdf

There is attenuation vs frequency with different codes.


I don't know how to read it.

For instance, how would a 1MHz signal respond to a particular DAC setting?

[tggzzz]

For instance, how would a 1MHz signal respond to a particular DAC setting?

Anywhere between -45dB and -5dB, depending on the bits set.

The way to read it is that you will get the output you "expect" at up to ~1kHz with all bits off, or to ~300kHz with all bits on. In other words, the frequency response depends on the DAC setting.

[CopperCone]

I get it.

So if the signal was gained after, it would still work

[T3sl4co1l]

"Work" in what sense?

If you need variable gain at more than 100kHz, a VGA, OTA or mixer may be a better option.

Tim

[CopperCone]

Im working on a testbed that can work with various signals from 1khz to 1 Mhz. I want ppm stability on the attenuation and repeatability. The nature of a dac attenuator allows it to be very temperature stable. It looks like i need a gain stage with a relay controlled resistor bank to implement.

I have seen a mixer used in a high frequency function generator before. I need to look at vga and ota. The thing is its meant to be a trim to scale a signal. A mixer can do that if connected to a dac. I think variable gain amps have very coarse steps. No idea what a ota is.

This is meant to replace what would need a bank of potentiometers. I want full digital calibration capability.

[T3sl4co1l]

Nice thing about potentiometers is they work up to low GHz (for the small, single turn, SMT ones with simple tracks).

If you need ppm with respect to code vs. frequency, you're probably not going to get it with this part even at 10kHz.  It's not clear if that's quite the spec you need.

OTA = Operational Transconductance Amplifier, typically used as the core of a VGA or (active, single or double balanced) mixer.  An LM13700 wouldn't give you the bandwidth here, but there are faster types.  The main downside is temperature sensitivity, usually compensated with a PTAT thermistor (or sensor + lookup table).  You'd probably not get ppm stability that way.

AFAIK, the traditional way these days is digital scaling, into a DAC with enough bits and sample rate to do the job.  Analog multipliers have always sucked, there just wasn't anything better until the last few decades (depending on what kind of thing you're talking about, from audio synth to RF).

Hell, what are you doing, anyway?  If that's low ppm's, it's not even meaningful without a 24 bit DAC, and 1MHz isn't very meaningful at 24 bits!

For a standalone adjustable attenuator sort of block, a chain of relays in binary dB steps (i.e., 40, 20, 10, etc.) is hard to beat.

Tim

[CopperCone]

I just want minimal drift with temp. I don't really care about how monotonous it is and stuff, so long I can set it up during calibration and have it stay that way.

DC offset is also important.

[mikerj]

Have a look at AD5543/AD5544 for multiplying DACs with more bandwidth.  If you don't really need 16bits then your choice will be larger for 10 or 12 bit DACs.

[CopperCone]

Those dacs look good. If i just want an attenuator do i need four quadrant multiplication?  For a bipolar input

I think that 2 quadrant is ok for attenuation and four if you want bipolar swing from a single supply.

Since the dac only has one input the concept of four quadrant seems weird if the terminology is the same as analog multipliers

Am i right that i only need a 2 quadrant setup?

[T3sl4co1l]

Four quadrant means you can set the gain positive or negative, on a reference signal that is positive or negative.  An attenuator doesn't invert, so two quadrant will get you there.  (As long as it's the right two quadrants...)

Tim

[CopperCone]

What do you mean right two quadrants?

I want the input to be bipolar and the output to be bipolar

I am going to use the AD5543/AD5544  recommended here.

[RandallMcRee]

He means the "correct" two quadrants.

You might want to look at the this dac: http://www.analog.com/en/products/digital-to-analog-converters/ad5780.html#product-overview

It features
0.025 LSB long-term linearity error stability and
±0.018 ppm/°C gain error temperature coefficient

which seems to be a feature that you desire. (Was just looking at these DACs and this is unique).

Four-quadrant explanation:
http://www.ti.com/lit/ug/tidu031/tidu031.pdf
This guide has a nice picture! (That's how I learn best ).

[CopperCone]

I don't see any frequency specifications in that datasheet unfortunately.

And yes, that app note cleared it up quite nicely.

[RandallMcRee]

Oh, right. I see now that AD "precision" DACs selector does not even have frequency response as a criterion, so the data sheets reflect that, too, I guess.

They do specify settling time which must equate to frequency response, indirectly.

Well, this leads me to believe that ppm stability and 1Mhz frequency must lead to a compromise in some respect. But, yeah, the AD5543 looks best from a speed point of view.

[CopperCone]

What exactly happens when its getting to its 3db point. Do you lose monotonacity? Get resonances that toggle ? More distortion ?

Im wondering if there are side effects from the extra circuitry used for the extra bits so that it might be better to use a part with less bits for better high frequency performance..

[RandallMcRee]

My understanding is that the settling time is the issue....as you have more bits in the DAC there are more switches and necessarily longer settling times to reach the stated resolution.

So, yes, fewer bits generally equates to better frequency response.