12

Please look at the schematic below. This is a very simple resistive adder that works fine with any standard ! (TTL,CMOS,...) or any arbitrary voltage that is fed into it. On the other hand as there is no active component in it, it is extremely fast. It is just made of a few resistors , so it is very cheap. On the other hand there is no limitation for number of input bits ( it can be easily expanded to 32, 64 or hundreds of bits ).

So, Why should we need DAC ICs? I am looking for a 32-bit high frequency DAC. Such devices are not found easily and also if found, they are rather expensive. I mean What is the advantage that I should pay for to find such devices? I think there must be some advantage that they worth to be bought. The only thing that I can think about is their inherent amplification ( for example TTL -> 10V or so), but this goal is simply achievable with any kind of amplification.

enter image description here

Aug
  • 1,561
  • 6
  • 28
  • 63
  • 12
    *"... there is no limitation on the number of bits".* Wrong. Do the error analisys. Think about it. How accurate do those resistor need to be to support your hypothetical 32 bit number? 1% resistors aren't even good to 7 bits. – Olin Lathrop Nov 04 '13 at 13:15

3 Answers3

19

What you have right there is what is known in the field as an R-2R DAC, one of the many different kinds of often employed digital-to-analog converter topologies. You have answered your own question: why do we need DACs when we have this DAC topology? Because it is a DAC!

R-2R DACs purely by themselves are not great as a general purpose digital to analog converter. The output impedance of an R-2R DAC is very high, which means that the bandwidth will quickly be very limited. Even a few tens of picofarads capacitance on the output will reduce effective bandwidth and increase settling time to the MHz region. And this is equally true if you buffer the output with an opamp follower - well-trimmed opamps don't come in sub-pF input capacitances, and reducing the R-2R ladder resistances quickly increases power consumption to the point where it is unacceptably high. Don't get me wrong, there are super high bandwidth R-2R DACs on the market, but these are the kinds of chips that you find in arbitrary waveform generators in some scopes, and they have a bit heatsink and fan on top of them.

There are other tradeoffs you can do with other DAC topologies. For instance, delta-sigma DACs do not to have a precision buffer output opamp and thus can be extended to very high bit depths (24-32 bit), where R-2R - because of the output buffering criterion - seldomly surpasses 12-bit. Successive approximation is another topology that is used, which inherently has a sample-and-hold on the output that can be driven with extremely low impedance (the same reason why conversely SAR ADCs can have a very high input impedance).

user36129
  • 8,268
  • 29
  • 37
  • 8
    The big problem with an R-2R is that for beyond a few bits the resistors need to be matched more accurately than available resistors, and the driving digital signals must also have the very precisely matched properties. Both make an R-2R with discrete components unpractical beyond let's say 8 bits. – Wouter van Ooijen Nov 04 '13 at 10:53
  • Thanks for very very good discussion! I never thought about the stray capacitance of R-2R design. So, what should I do for my 32-bit , 200MHz data? I can not find any suitable DAC in the market. – Aug Nov 04 '13 at 10:59
  • 3
    @WoutervanOoijen: actually, R2R ladders are very good at generating signals with lots of bits without any resistor matching. Even with the worst resistors in the world the code will still always fundamentally be monotonic (per bit). You get lots and lots of resolution, but because of the aforementioned problems with matching you get horrible precision and accuracy. This is why 16/24-bit audio DACs do exist in the R2R variant (also for their excellent low noise characteristics), but nobody in their right mind would use it as a precision DAC. – user36129 Nov 04 '13 at 11:03
  • 1
    @Aug. What are you trying to accomplish at 200MHz? Knowing might help us get you to a good answer – Scott Seidman Nov 04 '13 at 11:10
  • @ScottSeidman I have designed a signal generator, based on FPGA. The output is up to 70MHz and the FPGA ( Xilinx, Spartan3) works at 200MHz. It makes digital signal from a Lookup table and in 32bit resolution, I need a 32-bit DAC. – Aug Nov 04 '13 at 11:19
  • 4
    Give up on 32 bits, because that corresponds to 192dB dynamic analog range, and you can find plentiful "baseband" DACs: http://www.analog.com/en/digital-to-analog-converters/high-speed-da-converters/ad9747/products/product.html – pjc50 Nov 04 '13 at 11:21
  • 3
    It is **not true** that R-2R DACs are "inherently monotonic" regardless of resistor matching. It is trivial to construct a counterexample. Also, the output imepdeance is exactly R. – Dave Tweed Nov 04 '13 at 12:21
  • 32 bit resolution corresponds to approximately 250pV resolution for a 1V maximum amplitude. That is picovolt, 10^-12. – jippie Nov 04 '13 at 13:14
  • @user36129 What if I use small resistors (for example R=10Ω , assuming 10-bit digital inputs are in 3.3v and can send the needed 330mA )? Now the output impedance would be small and with 1pF stray capacitance, it should go well for 1.5GHz. – Aug Nov 04 '13 at 14:00
  • The assumption that you can turn a 330mA driver on and off at 1.5GHz is one I would challenge - you need a substantial MOSFET there - and at the same time have an output cap of 1pF. Also I notice you have accepted the goalpost moving to 10 bits rather than 32, at which point the DACs become plentiful. – pjc50 Nov 04 '13 at 14:58
  • 3
    http://www.analog.com/library/analogdialogue/archives/44-04/ad5791.pdf Analog app note on the difficulties of achieving 20 meaningful bits, at only 1MHz. Yes, it's an R-2R ladder - on a chip, with careful factory process control and trimming. – pjc50 Nov 04 '13 at 15:03
  • @pjc50 Yes! as jippie and you said it was a very optimistic assumption to use 32-bits! I looked at AD9850 datasheet and saw they use 10-bit DAC also. I think I should revise to 10-bits. But most DACs at that frequency are serial-DACs and my design is made for parallel. – Aug Nov 04 '13 at 15:53
4

What you have there is called an R2R-resistor ladder. The ICs you can buy also have such a network internally, but because it is integrated, it is much easier to guarantee the accuracy. See the Wikipedia entry as to why it is so important to have exact resistor values. I'd say it's almost impossible to achieve the accuracy of integrated circuits with discrete hardware.

Also many DACs have serial interfaces, so you don't need as many pins from your MCU to use them.

Geier
  • 504
  • 4
  • 13
1
  • Size vs bits of resolution (depends on the architecture)
  • Power comsumption (passive, also depends on the architecture)
  • Load effects (input / output impedances)
  • Quantization steps level dependant on devices connected (see load effects)
  • Precision
  • Accuracy
  • Noise
  • Bandwidth dependant on devices connected (see load effects)
  • Too many critical components ...
jsrmalvarez
  • 164
  • 4
  • This would be better if prefaced with something that answered the question. (i.e. What is this list for? Reasons for having DACs? Reasons R2R networks are not ideal for all situations?) – JYelton Nov 04 '13 at 18:15