# LM10 Reference...

Discussion in 'Electronic Design' started by Rich, Mar 24, 2006.

1. ### RichGuest

Hello,

1) I'd like to use the opamp portion of an LM10 as a comparator and the
reference portion to set the Vp input for the comparator. Vn will be
compared to the reference at Vp. Now, I'd like to adjust the input to Vp
of the opamp to a factor of 200mV (x * 200mV). So I think all I have to do
is use the reference feedback from pin-8 to pin-1 (R2) and tie pin-8 to GND
through R1. I'm using the pin out for the DIN package. I think this just
sets the gain for the reference output to 200mV * (R1+R2)/R1...just like a
normal non-inverting amp. Is that correct?

2) Also, for the comparator portion, it's my understanding that what makes a
comparator unique is that there is no feedback and the opamp will not
operate linearly. Thus, when Vp - Vn > 0 you get V+ (or near it) and when
Vp - Vn < 0 you get V- (or near it). That said, I've come across some
circuits lately with a very large resistor in the feedback path of what I
surmise is a comparator (1-10 MOhm). What would be a good reason for this?
I think the value is so high that VERY little current will flow; however,
I'm curious as to the practice.

Thanks!

2. ### Spehro PefhanyGuest

Probably POSTITIVE feedback to give the comparator a taste of
hysteresis. Otherwise the output could bounce around from noise or
oscillate for differential inputs that are close to zero. By fiddling
with the feedback you can achieve a circuit that does something like
go low for inputs > 2.0V and go high for inputs < 1.0V, and maintain
the previous state for voltages from 1.0 to 2.0V.

^
|
Vout

Vh ---------+----->------+
| |
| |
| |
| |
Vl +-----<------+------------

0 1.0V 2.0V

-----> Vin

Often much smaller amounts of hysteresis are used, of course.

Best regards,
Spehro Pefhany

3. ### Guest

As Spehro has said, this will work. Keep in mind that an op amp is
never a very good comparator - compators are designed to spend most of
their time saturated, and to come out of saturation very quickly, while
op amps are designed on the basis that they will spend all the time
un-saturated, so that when they do saturate, the designer hasn't gone
to any troulbe to prevent the accumulation of stored charge in
transistor bases in the ampliifer chain. I've seen horror stories about
741 amplifiers exhibiting propapagation delays of 500usec when used as
comparators.
As Sphero says, this 10M resistor is almost certainly providng a bare
minimum of positive feedback/hysterisis to prent the comapator
oscillating near threshold.

4. ### Jim ThompsonGuest

On 24 Mar 2006 02:36:09 -0800, wrote:

[snip]
[snip]

But it's not due to _device_ storage, it's due to the compensation cap
being charged against a rail and the very small currents available to
discharge it.

...Jim Thompson

5. ### Guest

Another misconception bites the dust. Thanks Jim.

6. ### Jim ThompsonGuest

Yep. It takes a mighty big device, like power BJT's, to store for a
microsecond.

...Jim Thompson

7. ### RichGuest

Well I've been RTFingM today...probably dangerous now...haha. So if I want
bias current (~20nA) and multiply by the large feedback resistance? Will
this be the amount of hysteresis I will have? Will it center around Vref
input because when the opamp/comparator is at Vcc I sink 20nA back to Vp
and when it's at Vee I source 20nA from Vp to the output? Ex: In my case,
Vref=1.7Vdc (8.5 * 200mV) so will I center at 1.7 - hyst when the output is
at Vee and at 1.7 + hyst when the output is at Vcc?

Is it better to use POSITIVE feedback to the reference input (Vp) of the
opamp OR to use NEGATIVE feedback to the (Vn) input? And most importantly,
why? I like the idea of using the Vp input in this case because it change
the reference value (1.7) slightly depending on whether the opamp is
saturated at Vcc or Vee. That said, I have no other technical merits to
offer as to why POSITIVE feedback would be better.

Also, once I add a large resistor I think I need to try and match the dc
resistance as seen by the opamp inputs. That large resistor has an impact
even if the input bias current is only 20nA! So let's say thst R1=1.87k
and R2=10k and I use 1MOhm for the feedback hysteresis. To match the dc
resistance seen by the opamp inputs I think I need to add ~1.5k from the
reference output to Vp, sound right?

And finally, this would probably be a whole lot easier for those reading if
I showed an ASCII schematic! What tool is available for this? Any linux
compatible alternatives? I was about to use VIm but was lazy!

Thanks!

8. ### Spehro PefhanyGuest

No, you usually want to use resistors of low enough value (or matched
well enough) that the input bias current doesn't have much effect.
It's not known very well, nor is it very stable with temperature.

Suppose you have the output swinging from 0 to 5V, with a 1K resistor
on the non-inverting input ( to ground or to a low-impedance source)
and a 1M feedback resistor. That means you'll have 5mV of hysteresis.
5V * 1K/(1K + 1M) = 5.0mV

Positive feedback gives you hysteresis, negative feedback gives you
gain that is lower than the open loop gain (or maybe oscillation, but
that's another story).
I don't know about your exact circuit, but 20nA * 1.5K is only 30uV,
which isn't worth worrying about in most cases, unless you're working
with really low-level DC signals (eg. from an R, S or B
thermocouple). Vos on this part is 2,000uV with a drift that isn't
specified, but typically is 2-5uV/K, which isn't great.
You can try Andy's (Andreas Weber's) AACircuit from here:
www.tech-chat.de

Best regards,
Spehro Pefhany

9. ### RichGuest

+-------10k-------+ +--------1MOhm-------+
| | | |
| | | +14Vdc |
| +----------+ | | | |
| |lm10 ref | | | | |
+---+---+Vn | | | +----+-----+ |
| | | | | |lm10 opamp| |
| | | | | | | |
| | +--+---1.5k------+--+Vp | |
| | | | | |
| |bal | | out+------+-------0
| | | | | Vout
| +----------+ +----+Vn |
1.33k | | |
| | +----+-----+
| | |
| battery>>> 0------10k-------+ |
| input | |
| | |
| 1.87k |
| | |
| | |
+---------------------------------+---------+
GND

NOTES:
1) Added 1.5k because Vn of opamp part "sees" 1/(1/10k + 1/1.87k) or 1.575k
2) The battery input will be ~10.8 when near 1.7 (Vref compare value)
3) 10k and 1.33k create 8.5 gain >>> 200 mV * 8.5 >>> Vref = 1.7
4) Added 1Mohm for some hysteresis...Now here's where I'm still a bit lost.
So my output swings from 0 to 14 Vdc. So the only low impedance path I
see is through 1.5k + 10k + 1.33k >>> 12.83k. So can I expect hysteresis
of 14 * [12.83/(12.83 + 1M)] = 177 mV? So the Vp input varies from an
ideal of 1.7 when Vout is ~0 to (1.7 + 0.177) when Vout is 14?
That makes better sense now. Thanks.
I will use for Windows and try it with wine in Linux!

10. ### John LarkinGuest

But if you're in no hurry, an opamp makes a nice comparator that won't
oscillate.

Actually, adding external hysteresis to a comparator is tricky. Most
comparators have multiple, fast gain stages so have a lot of delay
relative to output risetime. So it's entirely possible that a fast,
noisy signal will send bubbles of noise through the gain chain
*before* the output changes enough to kick in the hysteresis. So you
can propagate multiple edges through, and the hysteresis doesn't help
much. This can be unholy hell slicing something like a DDS output.

An opamp, usually having a single dominant pole, doesn't have this
problem.

A few comparators have internal hysteresis, a nice idea.

John

11. ### Jim ThompsonGuest

As in my venerable MC1560/61 design... designed in the mid '60's,
fastest on earth till almost present time.

(Latch feature was actually done using increased/insurmountable
hysteresis.)

...Jim Thompson

12. ### Phil HobbsGuest

Brr. I've never seen that one, but it sounds like huge fun to debug.
If you put at least a second-order filter on the DDS output, the slope
of the output won't change immediately due to a narrow glitch or
settling transient, the way it does with a first-order filter. Even two
cascaded RCs are usually enough for this. The order of the filter is
sometimes as important as the bandwidth.
Adding hysteresis by positive feedback also tends to put sharp-edged
junk back out the + inputs. The 1560 and MAX900 latches are in the
input structure, so you can put an RC differentiator between the output
and the latch--which gives most of the benefits of hysteresis, puts
nothing back out the input, and gives you a nice guaranteed minimum
pulse width. (*)

Disadvantages are: it only works in one polarity; it'll chatter with a
signal that sits right at the threshold; and you have to give the RC
time to discharge between pulses, like a POR.

Cheers,

Phil Hobbs

(*) Latches in the output structure tend to misbehave when you do this,
because the propagation delay between latch and output can be less than
the rise time.

13. ### Jim ThompsonGuest

Neeeerp! The MC1650/51 (sorry about the part-number typo above)
hysteresis is in the level translation from input stage to output
stage... I figure I ought to know, I designed it. See MC1650 PDF on
my website and the patent, 3,638,041

...Jim Thompson

14. ### Phil HobbsGuest

Hmm. Must have changed in the last couple of years then.... Here's a
post of yours from the last time we discussed this point.... ;-)

Cheers,

Phil Hobbs
--

15. ### Jim ThompsonGuest

I should have said "reflects to the input". The hysteresis appears in
the collector loads of the first stage. See the PDF, posted in the
past year when I got around to it.

...Jim Thompson

16. ### John LarkinGuest

Our benchtop delay generator uses an AD9832 DDS as the reprate
generator, and the 5-pole elliptic filter recommended by ADI. We tried
various fast comparators, with various amounts of hysteresis, and got
nasty jitter or multiple edges at various frequencies. The best thing
turned out to be a Tiny Logic schmitt trigger.

The elliptic filter is maybe not the best choice. It's optimized for
frequency-domain behavior, but has a purely capacitive signal path
mode, which I guess transmits glitches. And, as noted, a fast
comparator can do all sorts of things before the external hysteresis
loop wakes up. The HC schmitt is dirt cheap and has a large amount of
real, unteasable hysteresis. Of course, you have to cheat on the DDS
output compliance spec, but life's never perfect.

Lately, in their datasheets and appnotes, ADI doesn't even mention the
fact that their DDS chips need hairy antialiasing filters. The filters
are more expensive and a lot bigger than the chips themselves.

of energy, and sometimes you can figure something out by asking "how
many bits of information could this component possibly store?"

Need coffee.

John

17. ### Guest

http://www.analog-innovations.com/SED/MC1650-DataSheet.pdf

says its worst case propagation delay can be as high as 5.7nsec - only
marginally faster than the almost equally venerable Am685 (Giles J N
and Searles 1972 Electronics 45 112-7) and slower than the Plessey
SP9685

http://www.datasheets.org.uk/search.php?q=SP9685&sType=part

which gives a worst case propagation delay of 3nsec at 25C, 4nsec over
the full temperature range. Plessey sent me a free sample around 1978,
so your "almost present time" would seem to be back in the late 1970's
....

offered 2.5nsec typical, 3.5nsec worst case and we used quite a few of
those around 1989/90.

Which earth do you think you are living on?

18. ### John LarkinGuest

Surely not the one where people are gracious, grateful, or polite.

John

19. ### Jim ThompsonGuest

Of course, at my age that's just yesterday ;-)
This one. However a lot of those faster devices were cantankerous as
hell to keep from oscillating... the MC1650/51 never was... and I
designed it ~1965.

...Jim Thompson

20. ### Guest

This is a known problem for regular two-pole Sallen and Key low-pass
filters. Adding a (buffered) passive single pole in front of the
two-pole stage can improve the behaviour at frequencies appreciably
above the 3dB frequency.

And I don't like elliptic filers for cleaning up DDS outputs - when
Frank Bemelman and I were looking at the problem, I got enthusiastic
about Bessel filters, This was about two and half years ago, and I
can't remember why I went that way - I think it had to do with ringing
on step edges.
The DDS chips have got a lot faster in recent years, and you only need
hairy filters if you want to use the DDS to synthesise frequencies up
close to the DDS clock rate.

If Analog Devices were recommending elliptic filters, the application
engineer probably ended up get a flea in the ear from a dissatisfied
customer - Bob Pease and Jim Williams aren't representative application
engineers, and while Analog Devices does have lots of good people,
nobody has ever worked out how to avoid hiring the occasional
plausiable idiot.