Stable 0.5 volt reference?

[Ken Finney]

Anyone have any ideas for a creating a very stable 0.5 VDC reference, with
an minimum of board space? The best idea so far has been a higher voltage
reference followed by a resistor divider and a buffer, but the stability
(less than 0.1%) isn't there.

 

[John Woodgate]

I read in sci.electronics.design that Ken Finney

Anyone have any ideas for a creating a very stable 0.5 VDC reference, with
an minimum of board space? The best idea so far has been a higher voltage
reference followed by a resistor divider and a buffer, but the stability
(less than 0.1%) isn't there.

Close the loop; use the output voltage to buck the precision reference
on which the original voltage source is based. The output voltage can
then be made dependent on resistor values only.

 

[Joerg]

Hi Ken,

A couple other options:

Pick a reference that has the desired precision and puts out an integer
multiple of 0.5V, for example a good 2.5V reference. Then use a resistor
array such as one of those 8 resistor SO packs. These still have too
much tolerance as absolute values but you may be able to obtain one
where the relative tolerance is within your spec. Only the ratios
matter. Employ this array to divide down to 0.5V.

Another option would be laser trimming. I am sure they have the facilies
to do that at Boeing but it may be too much of a hassle if the rest of
your circuitry doesn't require laser trimming.

Regards, Joerg

 

[Rene Tschaggelar]

Anyone have any ideas for a creating a very stable 0.5 VDC reference, with
an minimum of board space? The best idea so far has been a higher voltage
reference followed by a resistor divider and a buffer, but the stability
(less than 0.1%) isn't there.

Why whould a divided and buffered reference be less stable ?

Rene

 

[Spehro Pefhany]

Anyone have any ideas for a creating a very stable 0.5 VDC reference, with
an minimum of board space? The best idea so far has been a higher voltage
reference followed by a resistor divider and a buffer, but the stability
(less than 0.1%) isn't there.

Just use a good reference, precision resistors and buffer it with an
op-amp if necessary.

Best regards,
Spehro Pefhany

 

[Thomas Magma]

Use a 3.0 VDC and a 2.5VDC reference and use the potential difference
between the outputs. 3.0 - 2.5 = .5 VDC

 

[John Woodgate]

I read in sci.electronics.design that Thomas Magma

) about 'Stable 0.5 volt reference?', on Fri, 16 Jul 2004:
Use a 3.0 VDC and a 2.5VDC reference and use the potential difference
between the outputs. 3.0 - 2.5 = .5 VDC

This is not good advice. If both voltages have a tolerance of 0.1%, the
resultant voltage can be between 3.003 - 2.4975 = 0.5055 V and 2.997 -
2.5025 = 0.4945 V. This is a tolerance of 1.1% - 11 times larger.
Stability is degraded by the same factor.

 

[Tim Wescott]

Hi Ken,

A couple other options:

Pick a reference that has the desired precision and puts out an integer
multiple of 0.5V, for example a good 2.5V reference. Then use a resistor
array such as one of those 8 resistor SO packs. These still have too
much tolerance as absolute values but you may be able to obtain one
where the relative tolerance is within your spec. Only the ratios
matter. Employ this array to divide down to 0.5V.

Another option would be laser trimming. I am sure they have the facilies
to do that at Boeing but it may be too much of a hassle if the rest of
your circuitry doesn't require laser trimming.

Regards, Joerg

I sat next to a guy who tried to find someone that would guarantee the
ratiometric tolerance of those resistor packs -- no one would. Either
the process that they use just isn't like a semiconductor process, or
the resistor folks don't have a clue.

He finally found someone; they were $$$ but small. I can't remember the
name, but I'll inquire if folks really want to know.

 

[Thomas Magma]

Nice catch John. It sounded good when it popped into my mind. Didn't do the
math.
Sorry Ken.

 

[legg]

Anyone have any ideas for a creating a very stable 0.5 VDC reference, with
an minimum of board space? The best idea so far has been a higher voltage
reference followed by a resistor divider and a buffer, but the stability
(less than 0.1%) isn't there.

IC references are available with stability <20ppm for constant
temperature situations. (I assume accuracy isn't important, or there
would have been some decimal places in the voltage.)

Divider resistors can track with similar stability.

Buffer offsets are limited only by what you want to pay, but you could
use an integrated circuit like the LM10, with an internal reference
and op amp, with an external transistor in the loop to remove
load-sensitive self-heating.

Then you could intentionally regulate the IC temperature using the
self-heating effect and a PTC parallel load. This would give long term
stability similar to oven-heated references, if the mechanical
environment was stable.

Other effects have to be approached as needed.

If it had to be SMD, I'm sure the same principals could be applied to
other devices.

RL

 

[Fred Bartoli]

Anyone have any ideas for a creating a very stable 0.5 VDC reference, with
an minimum of board space? The best idea so far has been a higher voltage
reference followed by a resistor divider and a buffer, but the stability
(less than 0.1%) isn't there.

One idea I used to amplify a reference with an *accurate integer* ratio was
to PWM the feedback loop.
A fast low rdson cmos switch, a counter, an opamp, some Rs and Cs, et voila.
The ratio only depends on the switch/counter propagation time. When you want
better ratio stability, simply decrease the switching frequency.

This probably won't fit your minimum size criteria but will very easily
satisfy your 0.1% accuracy requirement at an unbeatable cost. You'll need a
suitable clock too.
In case you already have an FPGA or a CPLD on board, all you need is a
switch and the Rs/Cs and that may be OK for you.


Thanks,
Fred.

 

[ddwyer]

Close the loop; use the output voltage to buck the precision reference
on which the original voltage source is based. The output voltage can
then be made dependent on resistor values only.

Rely on accurate mk:space ratio of a divider.
Hence 2.5V square wave 1t 2.5V 4T ground (0V) then average (LPF) the
result.

 

[Joerg]

Hi Tim,

I sat next to a guy who tried to find someone that would guarantee the
ratiometric tolerance of those resistor packs -- no one would. Either
the process that they use just isn't like a semiconductor process, or
the resistor folks don't have a clue.

He finally found someone; they were $$$ but small. I can't remember
the name, but I'll inquire if folks really want to know.

These can be had at 0.1% precision but the end of life notice makes me
think there isn't much of a market for this stuff anymore:

http://www.calmicro.com/products/data/pdf/prn_100-110.pdf

There are or at least used to be several other companies but nowadays
most people can live with references of standard bandgap value or
whatever is on the buffet tables of the big manufacturers. The rest is
done with ADCs and DACs of sufficient precision. Depending on what Ken
wants to do in his design a DAC solution may be yet another option for him.

Regards, Joerg

 

[Ken Smith]

I read in sci.electronics.design that Ken Finney

Close the loop; use the output voltage to buck the precision reference
on which the original voltage source is based. The output voltage can
then be made dependent on resistor values only.

I don't see how you do this. Surely, some voltage source or current
source must show up in the final equations.


I do agree that a closed loop system can reduce the error to the error of
the basic reference and the resistor divider.

 

[Ken Smith]

Why whould a divided and buffered reference be less stable ?

You have the error in the basic reference, the error in the divider and
the offset voltage in the buffer. If you can get the buffer's offset to
drop out, you do somewhat better.

 

[Ken Smith]

Anyone have any ideas for a creating a very stable 0.5 VDC reference, with
an minimum of board space? The best idea so far has been a higher voltage
reference followed by a resistor divider and a buffer, but the stability
(less than 0.1%) isn't there.

How about this:

Use some analog switches to switch capacitors in series onto a higher
voltage reference then into parallel to make the 0.5V. The advantage is
that compontent ratios don't come into it.

 

[Spehro Pefhany]

You have the error in the basic reference, the error in the divider and
the offset voltage in the buffer. If you can get the buffer's offset to
drop out, you do somewhat better.

Where would you get a buffer amplifier that bad? For stability, offset
doesn't matter, it's *change* in offset (TCVos and change with time).

Best regards,
Spehro Pefhany

 

[Winfield Hill]

Ken Smith wrote...

Use some analog switches to switch capacitors in series onto a higher
voltage reference then into parallel to make the 0.5V. The advantage
is that compontent ratios don't come into it.

Capacitance ratios certainly do, as well as switch charge coupling,
and output-amplifier offset voltages, natch. A useful technique
inside a cmos IC perhaps...

Thanks,
- Win

(email: use hill_at_rowland-dot-org for now)

 

[Winfield Hill]

Spehro Pefhany wrote...

Where would you get a buffer amplifier that bad? For stability, offset
doesn't matter, it's *change* in offset (TCVos and change with time).

Right.

Whether in a "buffer" amp or an inverting opamp, very low offset voltage
and even lower drift are easily had (e.g. Cirrus Logic cs3301). In this
500mV app, offset voltage is indistinguishable from absolute error in the
resistors, and would be trimmed out during cal in the same manner. Using
available 1ppm/C resistors and the above 0.05uV/C opamp, one could easily
make a 500mV reference with performance no more than 0.002% worse than
the initial voltage reference, over a operating 10C range.

Degrading to commonplace 50ppm/C resistors would suffice to substantially
exceed Ken Finney's miserable 0.1% claim. Or use 5ppm/C parts for 0.01%.

Thanks,
- Win

(email: use hill_at_rowland-dot-org for now)

 

[Fred Bloggs]

Where would you get a buffer amplifier that bad? For stability, offset
doesn't matter, it's *change* in offset (TCVos and change with time).

The main thing that *doesn't matter" is answering the asswipe OP's post!
The idiot probably "requires" a 0.5V reference because he's too dumb to
figure out how to boost a button cell to working voltage range. USENET
is a goddammed sewer.