AD9850 performance

[Garrett Mace]

I can find lots of projects designed around the AD9850, but haven't
seen one that actually compared the output with that of a commercial
function generator. Anybody know how well it stacks up against, say, a
decent Agilent model? Only on the basis of signal quality...not the
usability or special modes available on a "real" function generator.
It would be nice to see pairs of sine, triangle, and square waves on
the same scope, at several frequencies from about 50Hz to 50MHz. If
you want to get really fancy, perhaps some spectrum analysis at the
same frequencies.

 

[John Larkin]

I can find lots of projects designed around the AD9850, but haven't
seen one that actually compared the output with that of a commercial
function generator. Anybody know how well it stacks up against, say, a
decent Agilent model? Only on the basis of signal quality...not the
usability or special modes available on a "real" function generator.
It would be nice to see pairs of sine, triangle, and square waves on
the same scope, at several frequencies from about 50Hz to 50MHz. If
you want to get really fancy, perhaps some spectrum analysis at the
same frequencies.

But the AD9850 only makes sine waves.

John

 

[Rene Tschaggelar]

I can find lots of projects designed around the AD9850, but haven't
seen one that actually compared the output with that of a commercial
function generator. Anybody know how well it stacks up against, say, a
decent Agilent model? Only on the basis of signal quality...not the
usability or special modes available on a "real" function generator.
It would be nice to see pairs of sine, triangle, and square waves on
the same scope, at several frequencies from about 50Hz to 50MHz. If
you want to get really fancy, perhaps some spectrum analysis at the
same frequencies.

What is visible on a scope may not be the obvious stuff.
As already mentioned, it only does sinewaves.
The frequency accuracy is dependent on your oscillator.
The amplitude is not really adjustable.
Well with one resistor ... meaning to get a decent generator
you'll need a digital attenuator.
The phase noise is not that bad, see page 6.


Rene

 

[Robert Lacoste]

I can find lots of projects designed around the AD9850, but haven't
seen one that actually compared the output with that of a commercial
function generator. Anybody know how well it stacks up against, say, a
decent Agilent model? Only on the basis of signal quality... (...)

I never used the AD9850, but I do have experienced the AD9852. Basically the
answer is : "The signal quality can be very good as long as your design is
very good !". More specificaly the first thing is to design a very good
output low pass filter. I mean fully shielded, filtered power supply, etc.
On my first design I wanted to use both the sinus output and a square output
(using an on-chip comparator that is in the 9852), and the result was more
than bad : heavy high frequency noises on the sinus output, just because I
was routing back my nicely filtered sinus output to a noisy digital
environment. Then I suppresed this feature, and installed a clean low pass
filter in a two-cans shielded box with proper power supply filtering, with
the output amplifier in the second can of the same shield, and then I got a
signal as clean as on the documentation provided by AD...

Hope's that will be helpful,

Friendy yours,

 

[Garrett Mace]

But the AD9850 only makes sine waves.

John

Not strictly true, it can also produce square waves. I got carried away and
wasn't really looking for triangle waves in the product spec.

And sine and square waves are 99% of most people need anyway. The use here
is for a fast-ish function generator, 50 or 60 megahertz; sine and square
are pretty much what hobbyists use most anyway.

Anyway, if I ever need triangle waves I know how to make a halfway decent
auto-centering square-to-triangle integrator with gain control, as long as I
don't need more than 200khz or so.

 

[Garrett Mace]

What is visible on a scope may not be the obvious stuff.
As already mentioned, it only does sinewaves.
The frequency accuracy is dependent on your oscillator.
The amplitude is not really adjustable.
Well with one resistor ... meaning to get a decent generator
you'll need a digital attenuator.
The phase noise is not that bad, see page 6.

Can do square waves too...

For amplitude, I was going to use a variable gain amp anyway. Also want to
add in some offset control.

 

[Richard Hosking]

FRom memory the 9850 has a 10 bit DAC, while the 9852 is a 12 bit
device, with corresponding improvement in spurs. Phase noise is nearly
as good as the clock you use, with a penalty for using the PLL clock
multiplier in the 9852/4
I found the 9852/4 gave very good performance on a spectrum analyser for
most frequencies. However spur amplitude can vary dramatically with even
a small change in output frequency. This can be predicted though it is
complex and I havent tried to do it. I doubt you would get useful
information from a CRO unless there were serious problems - the output
is a very clean looking sine wave. I was able to get 1 Hz to 70 MHz with
the board at http://members.iinet.net.au/~richardh/VK6BRO.htm
and the Atmel driver board.

I am looking to do a board using the new 9951/9954 series - has anyone
done this yet? I would be keen to get a colaborative group going as I
dont have the time for these things nowadays! It should be possible to
get 160 MHz out directly, with much lower current drain than the 9854,
though the supply is 1.8V so I guess you would have to use a SMPS to get
the power savings from a 12V supply. It is a 14 bit DAC, so spurs are
better still. Anyone interested?

Richard

 

[John Larkin]

Not strictly true, it can also produce square waves. I got carried away and
wasn't really looking for triangle waves in the product spec.

OK, but you have to properly lowpass filter the DDS output and then
square it up with a comparator... just like any other sinewave source.
DDS chips don't make usable square waves all by themselves. The
filter/comparator thing can be a little tricky over a wide frequency
range, too.

John