TMS320F2812 ADC performance

Hi Everyone

I have two implementations of the F281x family, one is a 2811 on a board
that we thought we'd laid out exactly according to TI's recommendations, the
other a 2812 EzDSP board. On both of these we see a lot of dispersion (about
10 counts) when sampling a constant input signal. On the EzDSP the input is
provided by a potential divider between the analog VDD supply the analogue
ground, decoupled by a large tantalum cap, with a theoretical cutoff
frequency of 70Hz.

Is this normal performance, or is there a wicked layout trick that can be
used to improve it?

Cheers
Geoff

in eggy peggy:

gmeggyorteggyimereggyatfeggyrreggyareggyideggyoteggyit

"Geoffrey Mortimer" <me@privacy.net> wrote in message
news:3ss3fcFq1a0fU1@individual.net...
> Hi Everyone
>
> I have two implementations of the F281x family, one is a 2811 on a board
> that we thought we'd laid out exactly according to TI's recommendations,
the
> other a 2812 EzDSP board. On both of these we see a lot of dispersion
(about
> 10 counts) when sampling a constant input signal. On the EzDSP the input
is
> provided by a potential divider between the analog VDD supply the analogue
> ground, decoupled by a large tantalum cap, with a theoretical cutoff
> frequency of 70Hz.
>
> Is this normal performance, or is there a wicked layout trick that can be
> used to improve it?
>
> Cheers
> Geoff
>
> in eggy peggy:
>
> gmeggyorteggyimereggyatfeggyrreggyareggyideggyoteggyit
>
>

I found the same with 2810.. basically i came to the conclusion
that its just a crock of sh*t.. i now use an external analog devices part..
problem gone.. lol

Hello Geoff,

> 
> I have two implementations of the F281x family, one is a 2811 on a board
> that we thought we'd laid out exactly according to TI's recommendations, the
> other a 2812 EzDSP board. On both of these we see a lot of dispersion (about
> 10 counts) when sampling a constant input signal. On the EzDSP the input is
> provided by a potential divider between the analog VDD supply the analogue
> ground, decoupled by a large tantalum cap, with a theoretical cutoff
> frequency of 70Hz.
> 
> Is this normal performance, or is there a wicked layout trick that can be
> used to improve it?
> 

Nothing wicked, I just have never seen any split ground architecture 
work in my 20 years in the trade. I always use a single solid ground 
plane no matter what the app notes or the folks back at the university 
told us.

Here is one way to test: If you want to see lower frequencies run a coax 
directly to the ADC input pin of the DSP and ground it right next to the 
chip. Plug it into a scope and switch that to Hi-Z (usually 1M). Look at 
what's there. If it's higher frequencies or your driving circuit can't 
stomach the coax's capacitive load (which would not be a good sign...) 
use a suitable resistive divider made from SMT resistors right at that 
pin and use the 50ohm setting on the scope. If the scope isn't sensitive 
enough you may have to use a spectrum analyzer here.

I mostly use RG174 coax because it is thin and pliable. Resist the 
temptation to just hook a regular scope probe and ground clip to the 
input. The loop of the ground clip wire can pick up lots of noise.

If you see much less crud on the scope or analyzer than on your data the 
ADC most likely isn't that great. If you see lots of bad stuff on the 
screen it's time to debug the board areound the DSP. Another quick test 
   is to connect a large (but physically small) ceramic cap from the ADC 
input pin to ground. It should get really quiet on the data output. Let 
us know what you see.

Most on-chip ADC need to be driven from a clean low impedance source.

Regards, Joerg

http://www.analogconsultants.com

"Joerg" <notthisjoergsch@removethispacbell.net> wrote in message
news:FZ6af.8910$7h7.4292@newssvr21.news.prodigy.com...
> Hello Geoff,
>
> >
> > I have two implementations of the F281x family, one is a 2811 on a board
> > that we thought we'd laid out exactly according to TI's recommendations,
the
> > other a 2812 EzDSP board. On both of these we see a lot of dispersion
(about
> > 10 counts) when sampling a constant input signal. On the EzDSP the input
is
> > provided by a potential divider between the analog VDD supply the
analogue
> > ground, decoupled by a large tantalum cap, with a theoretical cutoff
> > frequency of 70Hz.
> >
> > Is this normal performance, or is there a wicked layout trick that can
be
> > used to improve it?
> >
>
> Nothing wicked, I just have never seen any split ground architecture
> work in my 20 years in the trade. I always use a single solid ground
> plane no matter what the app notes or the folks back at the university
> told us.
>
> Here is one way to test: If you want to see lower frequencies run a coax
> directly to the ADC input pin of the DSP and ground it right next to the
> chip. Plug it into a scope and switch that to Hi-Z (usually 1M). Look at
> what's there. If it's higher frequencies or your driving circuit can't
> stomach the coax's capacitive load (which would not be a good sign...)
> use a suitable resistive divider made from SMT resistors right at that
> pin and use the 50ohm setting on the scope. If the scope isn't sensitive
> enough you may have to use a spectrum analyzer here.
>
Hi Joerg, thanks for the reply

I like that resistive divider/coax trick - first saw it in a great book
entitled

"High speed Logic Design - The Art of Black Magic"

The idea, by the way, for other readers, is to have a source impedance of 50
Ohms driving the coax - so, for instance, if you want to restrict the load
on the driving circuit to about1k, use a divider made up of 950 Ohms in
series with 50 Ohms to ground. This makes a 20x probe the input impedance of
which is1k and the output impedance about 50 Ohms. It's a neat trick.

> I mostly use RG174 coax because it is thin and pliable. Resist the
> temptation to just hook a regular scope probe and ground clip to the
> input. The loop of the ground clip wire can pick up lots of noise.
>
> If you see much less crud on the scope or analyzer than on your data the
> ADC most likely isn't that great. If you see lots of bad stuff on the
> screen it's time to debug the board areound the DSP. Another quick test
>    is to connect a large (but physically small) ceramic cap from the ADC
> input pin to ground. It should get really quiet on the data output. Let
> us know what you see.
>
In the test circuit I have a 2.2uF 1206 tant between the input pin and
ground - in my experience these are quite good at the frequencies at which
ADCs operate. I'll try adding a 10n ceramic in parallel, but I don't hold
out much hope.

When I get the chance, I'll have a look with the scope ...

> Most on-chip ADC need to be driven from a clean low impedance source.
>
> Regards, Joerg
>
> http://www.analogconsultants.com

Geoffrey Mortimer wrote:

   ...

> I like that resistive divider/coax trick - first saw it in a great book
> entitled
> 
> "High speed Logic Design - The Art of Black Magic"
> 
> The idea, by the way, for other readers, is to have a source impedance of 50
> Ohms driving the coax - so, for instance, if you want to restrict the load
> on the driving circuit to about1k, use a divider made up of 950 Ohms in
> series with 50 Ohms to ground. This makes a 20x probe the input impedance of
> which is1k and the output impedance about 50 Ohms. It's a neat trick.

Note that with resistors in the same package, the shunt capacitances 
will be about equal. That makes the high-frequency ratio close to 1:1, 
peaking the highs. (The trim in a scope probe adjusts for that.) I leave 
it to your ingenuity and available space to get around that.

   ...

Jerry
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������

Hello Geoff,

> I like that resistive divider/coax trick - first saw it in a great book
> entitled
> 
> "High speed Logic Design - The Art of Black Magic"
> 

For anyone doing high speed digital or critical analog designs and not 
being 100% familiar with all the tricks yet this book is a "must have".


> The idea, by the way, for other readers, is to have a source impedance of 50
> Ohms driving the coax - so, for instance, if you want to restrict the load
> on the driving circuit to about1k, use a divider made up of 950 Ohms in
> series with 50 Ohms to ground. This makes a 20x probe the input impedance of
> which is1k and the output impedance about 50 Ohms. It's a neat trick.
> 

I must confess that I am not always too diligent about this. If a coax 
is well terminated at the other end (scope, analyzer) there won't be 
much of a reflection problem even if the source side isn't equal to Z. 
This is because then the scope won't reflect anything that could bounce 
back a 2nd time from the source side. In daily life that is also the 
reason why we don't get ghosting on TV if the antenna amp is a clean 
75ohms (the antenna never really is over the whole UHF range). In tight 
spots I sometimes just hang a series resistor up front.

Jerry's remark is important though. In case you have to watch a large 
frequency range, for example to diagnose fast digital transients or if 
you suspect noise getting in from a digital circuit, the spurious 
capacitive ratio of your divider has to be minded. Else the signal is 
already messed up before it begins its travels up the coax.

> 
> In the test circuit I have a 2.2uF 1206 tant between the input pin and
> ground - in my experience these are quite good at the frequencies at which
> ADCs operate. I'll try adding a 10n ceramic in parallel, but I don't hold
> out much hope.
> 

I don't know your frequency range of interest but tants aren't very 
suitable here. Use a ceramic. They come in surprisingly large values 
these days for a tiny 0805. I just had one in a design that was 4.7uF 
yet so small I that had to use tweezers (and my glasses to see it...).

If you still have separate grounds that are on the same DC potential try 
connecting them at several places, short connections. Done that a lot at 
clients and usually their jaws dropped when they looked at their FFT 
afterwards. Especially since the production cost for this change is zilch.

Regards, Joerg

http://www.analogconsultants.com