Hi,

I've been working on a VGA DAC for my upcoming CPLD graphics project. I wanted it to be more accurate than the usual quick-and-dirty resistor ladder arrangement, mostly as a fun challenge, as I have never worked with analog circuitry before.

In my research for a suitable opamp buffer, I came across the LT1260 triple amplifier which seems purpose made for driving RGB signals and is available in DIP, though a bit expensive at 7€ (https://nl.mouser.com/datasheet/2/609/1 ... 123073.pdf). So, I designed the following circuit around this IC. It uses pre-made 4-bit R2R ladders, with a multiturn potentiometer on each channel allowing me to calibrate the outputs to 0.7Vp-p as required by the VGA signal.

Even though I was only able to find 4-bit R2R ladders, the intent is to use this circuit with a RRRGGGBB palette, I'm only including the additional bits because they're free.

The color signals have proper 75ohm output impedance, and I added 75ohms (because it simplifies the BOM) source termination on HSYNC and VSYNC, though I'm not sure if that's useful. The gain and feedback resistors are 1.1k as recommended by the datasheet.

The negative power supply is generated by a MAX1044 (https://www.analog.com/media/en/technic ... AX1044.pdf) which is a charge pump, so probably not accurate enough for this? I'm struggling with how to generate this negative power nicely.

As mentioned, I'm not really an analog guy, and would love to hear any feedback you may have.



I also created a simulation on circuitjs

_________________
BB816 Computer YouTube series

_________________
http://WilsonMinesCo.com/ lots of 6502 resources
The "second front page" is http://wilsonminesco.com/links.html .
What's an additional VIA among friends, anyhow?

I wonder if you're over-thinking it...

So just as another data point, there exists a VGA adapter for the Raspberry Pi that's driven purely from the GPIO connector (by the GPU) that's nothing more than an R/2R network and it works remarkably well without any buffer amplifier, trim pots, etc. It's only 6 bits per channel but for the most part you'd never know it wasn't the full 24-bit RGB and despite passive components it run right past full HD resolutions too.

Search for Gert VGA666 for more details. (inc. schematic, etc.)

-Gordon

_________________
--
Gordon Henderson.
See my Ruby 6502 and 65816 SBC projects here: https://projects.drogon.net/ruby/

I'm not sure you need that negative voltage. As far as I know the VGA RGB signals are 0V to 0.7V. It's possible that the LT1260 can be run off +5V positive and 0V negative but I only think that from a quick scan of the datasheet - don't trust me until you've tested it.

You might not need the LT1260 at all. If you haven't seen them then James Sharman's videos are a pretty good reference: DAC Test - VGA from Scratch - Part 10.

Thanks! I came across this useful app note from Analog Devices showing the best ways to implement gain adjustment: https://www.analog.com/en/technical-art ... cuits.html. Based on that and your advice, I've now moved the trimmers.

For the charge pump, I was worried because it shows a significant drop in output voltage, 1V between 0mA and 20mA which is what the three opamps would typically use. However, I since realized I had already purchased a MAX739 a long time ago for a DRAM project, so I updated the power supply to be based on that. It is a regulator and should be more stable in the range I'm using it, although it needs a few more passives.

Now, that makes both ICs used on this board quite expensive, at 7€ each. Plus the trimmers and R2Rs, it is an expensive board. Maybe in the future I'll try to reduce costs, I know going to SMD will likely make this a fraction of the price, but for now I want to experiment.



Yes, I'm most definitely over-thinking it. I agree that a couple of resistors is all you need, but where is the fun in that! If I can get marginally better color reproduction and signal integrity, while learning something in the process, then I'll be happy.



Thanks! I watch James' videos religiously, and his approach is sound. What's missing from his presentation is the effect of current on the R2R ladder. If a bit of current flows through the cable, or you have signal integrity problems due to the impedance mismatch and not driving it at 75 ohms, then I would argue that it could generate errors comparable to the 1% tolerance of resistors, especially noticeable in 8 bits of resolution. I have not measured this though, so maybe I'm overstating it. Still, using an opamp buffer is an approach that should fix that and a learning opportunity for me!

On the power supply issue, they do recommend +/- 5V, but also say it could run with a single supply. I'm not sure how to interpret the datasheet to see how much error there would be at 0V in a single supply circuit? And I'm also not familiar at all with DC biasing and filtering to make the opamp work in a better range, maybe that's for a future learning project.

_________________
BB816 Computer YouTube series

For a charge pump I use the TC7660 which is pin compatible with the MAX1044. You can see it in the right hand side of this schematic. It comes in at around a US dollar making it a bit cheaper than the MAX1044. I'm also not familiar with DC biasing so I'd probably just use a charge pump for -5V and be done.

If you're willing to try a completely off-the-wall solution you could use a 30bit video DAC like the THS8136 or a 24bit video DAC like the ADV7125. Those will both handle the current for you (with appropriate 75Ω termination resistors). I'm using the '7125 but that comes with a few gotchas. Mostly that it's not produced anymore*. And also that both are capable** of encoding the sync signal in the green channel.

I'm not sure that the above is a solution in the spirit of your project though.

* available for around a US dollar on aliexpress if you're brave.
** but can be disabled.

Those are very interesting ICs, thanks!

To be honest though, if I end up going for a premade version of the DAC in the future, in a TQFP package, I might as well go digital and use the TFP410, similar to how 1bitsquared is doing with their PMOD DVI https://1bitsquared.com/products/pmod-d ... -interface. Though supply for this one is a bit delayed at the moment.

_________________
BB816 Computer YouTube series

A resistor DAC is not really quick-and-dirty IMO, it is pretty much a case of a simple solution that is also the right solution. I suppose you might consider the binary-ladder DAC quick and dirty, and the R-2R DAC the sophisticated solution.

For VGA (analog video), because the digital voltages are known, and the analog voltage range and impedance are also known, you can calculate the values you need for either resistor DAC type.

For a DAC you have are few parameters you are trying to achieve:

Accuracy - how well the same binary value produces the same analog voltage.
Precision - how closely the binary value can be represented by the analog voltage.
Resolution - how small of a change can be represented.
Linearity - how equal the analog change is between any two sequential binary input values.

The main problem for a binary resistor DAC is finding resistor values that are exact powers of two, i.e. 1K, 2K, 4K, 8K, etc.. Since resistors don't come in exact values like that, and their tolerance varies a lot, these kinds of DACs become difficult to make when the binary input is large, i.e. 6 or more bits per channel. For a small video DAC, like 3 or 4 bits per color, this kind of DAC will work great and is probably well outside of a human eye to tell the difference. Use 1% or 0.5% tolerance quality resistors and you can get nice video. I'll attach a photo of a 9-bit (RRRGGGBBB) binary resistor DAC.

For more bits, the R-2R ladder is well suited because you only need two different values of resistors which are easier to achieve.

Analog electronics are much harder than digital, and noise is harder than analog. If you really want to "up your DAC game", then the single most critical parameter is going to be your power supply voltages, and controlling noise.

A VGA signal is 0V to 0.7V full range, per color channel. For full 24-bit color, there are 8-bits per channel, which is 0.7 / 256, or 0.00273V per gradient change. So, the first thing you need is a power supply for your digital electronics that is accurate to around +/-0.00002V with ripple that is even smaller.

The amplifiers on the output of your DAC are fine if you need the drive, but they do not help with with parameters above. Actually, your amplifier will also need the low-noise power supplies, and its bandwidth and linearity will matter greatly.

Last, all this is moot for digital video, i.e. Display Port, DVI, and HDMI. And if you are driving analog VGA for a hobby project, then a 3-bit or 4-bit resistor DAC will produce a nice result that you will not be able to distinguish from anything else. IMO.

_________________
http://WilsonMinesCo.com/ lots of 6502 resources
The "second front page" is http://wilsonminesco.com/links.html .
What's an additional VIA among friends, anyhow?

Thank you both for very good feedback. I won't be able to reply for a few days but I'll be back!

_________________
BB816 Computer YouTube series

For a 3 bit DAC, exact powers of 2 are easy, as long as you have the middle value resisrtor to hand.

2 resistors in parallel will give you half the value.
2 reistors in series will give you double.

3 "perfect" values, at the expense of using 5 resistors per colour, instead of 3.

As pointed out above, the video output from a VGA port is 0.0 to 0.7v, as measured into 75 ohms built into the display. So open circuit, you should expect to see 1.4v at peak white (er, peak rgb).

Note that this means the black level is sat on the supply zero which might have implications if your local ground is insufficiently stiff - though with a bi-rail op-amp this shouldn't be a problem. Op-amps which work to 'ground' rarely actually make it without distortion as they approach the rail.

A PAL or NTSC composite video is raised 0.3v to include the sync pulses, so runs from 0.3v to 1.0v. In broadcast applications, exactly 0.3v is considered 'superblack' and is never (deliberately) transmitted; the lowest black is at 2% (0.014v/0.314v depending on component or composite).

Neil