Any D/A converter and amplifier for microwave frequencies?

[mheslep]

> the 4GHz clocks would never be exposed outside the chip

I don't understand what you mean. The first solution I found, just the chip, accepts external clocking and its evaluation board does just that.

The clock input for modern Intel CPU chip is the at the 'front side bus' frequency and is something like 233, 300, or 400 MHz. The chip takes that frequency and internally multiplies it by say, 17X, to get the GHz ratings you see them sold by. Thus the highest fundamental clock frequency running around on the PCB (mother board) in the 100's of MHz; the GHz clocks never leave the chip. One reason is that digital clocks in the GHz range are much more difficult to manage. So if you have in fact found a DAC that accepts a ~2GHz digital clock, I believe you'll find it difficult to deliver the clock to the DAC. The data sheet likely has notes on this, describing special layout procedures (perhaps strip line design is required as suggested by Chroot). Its highly unlikely that the usual inexpensive FR4 printed circuit will suffice. Also, I'm guessing that with a digital clock that high thermal management becomes highly problematic (power dissipated proportional to the clock rate) and you'll have to add heat sinks, etc. I think it highly unlikely you'll find a clock input all the way to 4GHz on any simple DAC as stated above; again that become more of a system than a chip and the cost will scale accordingly.

 

[f95toli]

>
How does it follow, from posting the link to the (non-existent) 2.3 GHz converter, that I do not listen to people? Am I supposed to listen to you, or obey you? :)

It is NOT a "2.3 GHz converter". It is a D/A converter that can handle inputs of 2.3 Gbps, which is why the bandwidth is 1GHz (which in this case means you can e.g. use to generate a signal between DC and 1 GHz, but not higher than that).
Now, this is still quite impressive; but it it is still nowhere near the performance you were asking for.

I think the point you are missing is that there is HUGE difference beween being able to handle PULSES at a rate of a few gigabits per second and handling SIGNALS with a BANDWITH of a few GHz; the latter is much, much more difficult.

 

[Ulysees]

Thank you mheslep. Just out if interest, do you have any thoughts how accurate a clock is? (irrespective of whether it's multiplied internally or not). I think it doesn't matter for my application if a sinewave has some very weak harmonics, what matters is that the fundamental is sharp, ie that the clock does not wobble. Can clocks be accurate?

 

[Ulysees]

> It is NOT a "2.3 GHz converter". It is a D/A converter that can handle inputs of 2.3 Gbps

Yes, didn't mean anything different. Thought it was obvious we meant samples per second, not Nyquist frequency or anything like that.

> Now, this is still quite impressive; but it it is still nowhere near the performance you were asking for.

Is it? There must be some filtering, maybe sinewaves you make with it are not so bad 5 times below the 1 GHz, as I just said above it's probably the fundamental that needs to be super-pure when looking for those supposed very-sharp resonances.

> I think the point you are missing is that there is HUGE difference beween being able to handle PULSES at a rate of a few gigabits per second and handling SIGNALS with a BANDWITH of a few GHz; the latter is much, much more difficult.

I've always been aware of this. I have a degree in Electronic Engineering from a top university in the UK, and it was a First. So feel free to go into the maths of it, it's all familiar.

 

[chroot]

Low-jitter 4 GHz clocks are indeed very difficult to generate and pass around a PCB. In most labs, you'd use a $50k frequency synthesizer to produce a highly-stable sine wave signal and use that for a clock directly, or pass it through some kind of a sine-to-square converter like an ECL gate. (The Maxim part appears to handle either sine or square clocks.)

The Maxim part has pretty good dynamic performance characteristics, with pretty high SFDR at low frequencies. I tried to look for the SFDR at higher frequencies, but they do not offer the datasheet to the public.

From Maxim's website:

This product is designed for a highly specific application. The detailed data sheet contains information about proprietary technology, and is only available to customers whose requirements closely match this application. To request the data sheet, please complete the following information detailing your requirements.

Unfortunately we are unable to respond to requests that do not meet our qualification criteria. If you do not receive a response within two business days, please select another part. Technical support for this product is NOT available from our normal Tech Support desk.

I'm guessing they won't actually sell you one, anyway.

- Warren

 

[Ulysees]

Can anyone guess what the highly specific application is for the Maxim part?

Also they don't say much about the evaluation board that comes with it, no mention of memory or anything. How is it likely to be driven from a pc?

 

[chroot]

I'm going to assume it's intended for military use, probably in a multi-band software radio. They make it pretty clear that it's intended for direct IF use in communications.

There's essentially no chance that it's meant to be driven by a PC -- not directly anyway. You do realize, I hope, that there's no way at all to actually get 2.3 billion 12-bit samples (almost 3.5 gigabytes of data per second) out of a PC. Normal PCs have only a fraction of the aggregate bus bandwidth necessary for moving this kind of data around, much less the external drivers and so on necessary to get it out onto a wire.

If I had to guess, I'd assume they have a programmable high-end FPGA on the demo board, which you probably CAN load with pattern data from a PC, but not in real-time. It's also highly probable that the board only provides a high-speed connector that allows you to connect it to a high-end pattern generator.

- Warren

 

[Ulysees]

> there's no way at all to actually get 2.3 billion 12-bit samples (almost 3.5 gigabytes of data per second) out of a PC.

Decent graphics cards from as far back as... 2006 came with PCI-express 16x which is up to 4 Gbytes/s. Here's one:

http://www.nvidia.com/object/7_series_techspecs.html
http://www.hothardware.com/articles/NVIDIA_GeForce_Go_7950_GTX_Preview1/?page=2

Do I win anything? :)

 

[chroot]

Just because one link in the chain can deliver 4 GB/s does not mean every link in the chain can. Good luck to you.

- Warren