Single vs cascading op amp
bob.jo...@gmail.com
stages of a signal rather than use a single stage if the need is to
use a single stage?
If each stage has gain G_i why can't we always just choose G_1*...*G_2
as the gain for the first stage assuming the op amp can handle it and
we can accurately find the resistors?
John Larkin
A single opamp will have a finite gain-bandwidth. And you generally
don't want to use all of it, because that will increase distortion.
So if the voltage gain times required bw is well below the opamp's
specified GBW, one opamp is usually enough.
There are other special cases, like not wanting a precision front-end
stage to dissipate a lot of power driving a load. Or wanting to use,
say, a zero-drift low-voltage amp, followed by something bipolar that
can swing lots of voltage.
John
Bob Eld
"Bob.Jo...@gmail.com" <bob.jo...@gmail.com> wrote in message
news:ac7736e1-df1f-470e...@r33g2000yqn.googlegroups.com...
Your question is not very clear. But, in general, each gain stage adds and
amplifies noise plus each stage has a finite bandwidth. Cascaded stages
reduce overall band width and increase noise Therfore you should use as few
stages as possible for a given situation.
Secondly if the stages are within a feedbagck structure like in an op-amp
the collective phase shift of the stages will add giving a high overall
phase shift. Since each stage produces 90 of phase degrees shift at the
limits of it's band width, two such stages will produce 180 degrees. If the
gain is above unity when the phase shift becomes 180 degrees, the amplifier
will oscillate or become unstable. For this reason most amplifiers and
op-amps are rarely designed with more than two internal gain stages. Three
or more gain stages are very difficult to stabilixe and keep from
oscillating.
John Larkin
wrote:
>
>"Bob.Jo...@gmail.com" <bob.jo...@gmail.com> wrote in message
>news:ac7736e1-df1f-470e...@r33g2000yqn.googlegroups.com...
>> Is there ever a time when it would be better to have multiple gain
>> stages of a signal rather than use a single stage if the need is to
>> use a single stage?
>>
>> If each stage has gain G_i why can't we always just choose G_1*...*G_2
>> as the gain for the first stage assuming the op amp can handle it and
>> we can accurately find the resistors?
>
>Your question is not very clear. But, in general, each gain stage adds and
>amplifies noise plus each stage has a finite bandwidth. Cascaded stages
>reduce overall band width and increase noise Therfore you should use as few
>stages as possible for a given situation.
If the first stage has significant gain, the noise of downstream
stages don't matter.
And more opamps increase available gain-bandwidth, not decrease it.
John
Tim Wescott
bob.jo...@gmail.com
<jjSNIPlar...@highTHISlandtechnology.com> wrote:
> On Mon, 6 Jul 2009 10:40:04 -0700, "Bob Eld" <nsmontas...@yahoo.com>
> wrote:
>
>
>
> >"Bob.Jones5...@gmail.com" <bob.jones5...@gmail.com> wrote in message
> >news:ac7736e1-df1f-470e...@r33g2000yqn.googlegroups.com...
> >> Is there ever a time when it would be better to have multiple gain
> >> stages of a signal rather than use a single stage if the need is to
> >> use a single stage?
>
> >> If each stage has gain G_i why can't we always just choose G_1*...*G_2
> >> as the gain for the first stage assuming the op amp can handle it and
> >> we can accurately find the resistors?
>
> >Your question is not very clear. But, in general, each gain stage adds and
> >amplifies noise plus each stage has a finite bandwidth. Cascaded stages
> >reduce overall band width and increase noise Therfore you should use as few
> >stages as possible for a given situation.
>
> If the first stage has significant gain, the noise of downstream
> stages don't matter.
>
> And more opamps increase available gain-bandwidth, not decrease it.
>
> John
The total gain would be the product of the individual gains and the
GBWP at the lowest frequency would be larger than one. But depending
on how you look at it you are either increasing the GBWP or decreasing
it.
For example, suppose you have two op amps with GBWP at 1Mhz and the
other at 10Mhz. The "total" GBWP would be somewhere inbetween 1Mhz and
10Mhz. But since the total GBWP is the product of the two we have
reduced the GBWP of the 2nd amp or increased that of the 1st amp.
or GBWP1 <= GBWP <= GBWP2
So your both right or both wrong depending on how you look at it ;) If
say, GBWP = 5Mhz then we have decreased the GBWP of the second op amp
from 10Mhz to 5Mhz while increasing the 1st from 1Mhz to 10Mhz.
Although, I think since the gain will drop off somewhat drastically
with frequency we have decreased the GBWP of the 2nd op amp much more
than we have increased the first. Also it increases the fall off of
the gain w.r.t to frequency as we approach the GBWP.
John Larkin
I was looking at using a number of identical opamps, running
closed-loop, to achieve some non-trivial target overall gain.
Suppose we need a gain of 100 overall, and we have a lot of 1 MHz
opamps around.
A single-stage design would have 100x all in one opamp, net bandwidth
10 KHz.
A two-amp cascade needs a gain of 10 per stage. Each stage is down 3
dB at 100 KHz, so the overall bandwidth is a bit less, 80K or
something like that.
Four opamps need gain per stage of 3.16, which is 316 KHz bw per
stage...
There's some derivation somewhere that figures out the maximum GBW you
can get from cascading identical amp stages. The gain per stage turns
out to be sqrt(e).
John
Jim Thompson
Cascading N identical stages of bandwidth W yields a net bandwidth
of...
Wnet = W*sqrt(2^(1/N)-1)
It is left as an exercise for the student to calculate net
gain-bandwidth ;-)
And input-referred noise (to satisfy Hobbs ;-)
...Jim Thompson
--
| James E.Thompson, P.E. | mens |
| Analog Innovations, Inc. | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| Phoenix, Arizona 85048 Skype: Contacts Only | |
| Voice:(480)460-2350 Fax: Available upon request | Brass Rat |
| E-mail Icon at http://www.analog-innovations.com | 1962 |
I love to cook with wine Sometimes I even put it in the food
George Herold
George Herold
Site.com> wrote:
> On Mon, 06 Jul 2009 17:19:39 -0700, John Larkin
>
>
>
>
>
>
> I love to cook with wine Sometimes I even put it in the food- Hide quoted text -
>
> - Show quoted text -
Thanks Jim, I didn't realize it dropped off so fast. 5 cascaded gain
stages and the BW is down by almost 1/3!
This merits more study!
George Herold
Electronworks.co.uk
"John Larkin" <jjSNIP...@highTHISlandtechnology.com> wrote in message
news:2ae4555j7aq037s94...@4ax.com...
John makes a good point - in multiple gain stages, it is advisable to have
the first stage with lots of gain (without running into your GBW
limitations). Getting the signal up as early as possible improves the signal
to noise ratio of the amplifier chain
--
Bill Naylor
www.electronworks.co.uk
Electronic Kits for Education and Fun
Robert Baer
amount is divided by gain of first stage.
John Larkin
Noise adds trig-wise (root of sum of squares) so if the first stage
has a gain of 10, the second stage only adds about 1% to the total
noise.
John
John Larkin
Squared.
John
George Herold
<jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
> On Tue, 7 Jul 2009 07:40:39 +0100, "Electronworks.co.uk"
>
>
>
>
>
>
> >"John Larkin" <jjSNIPlar...@highTHISlandtechnology.com> wrote in message
> >news:2ae4555j7aq037s94...@4ax.com...
> >> On Mon, 6 Jul 2009 10:40:04 -0700, "Bob Eld" <nsmontas...@yahoo.com>
> >> wrote:
>
> >>>"Bob.Jones5...@gmail.com" <bob.jones5...@gmail.com> wrote in message
> >>>news:ac7736e1-df1f-470e...@r33g2000yqn.googlegroups.com...
> >>>> Is there ever a time when it would be better to have multiple gain
> >>>> stages of a signal rather than use a single stage if the need is to
> >>>> use a single stage?
>
> >>>> If each stage has gain G_i why can't we always just choose G_1*...*G_2
> >>>> as the gain for the first stage assuming the op amp can handle it and
> >>>> we can accurately find the resistors?
>
> >>>Your question is not very clear. But, in general, each gain stage adds and
> >>>amplifies noise plus each stage has a finite bandwidth. Cascaded stages
> >>>reduce overall band width and increase noise Therfore you should use as
> >>>few
> >>>stages as possible for a given situation.
>
> >> If the first stage has significant gain, the noise of downstream
> >> stages don't matter.
>
> >> And more opamps increase available gain-bandwidth, not decrease it.
>
> >> John
>
> >John makes a good point - in multiple gain stages, it is advisable to have
> >the first stage with lots of gain (without running into your GBW
> >limitations). Getting the signal up as early as possible improves the signal
> >to noise ratio of the amplifier chain
>
> Noise adds trig-wise (root of sum of squares) so if the first stage
> has a gain of 10, the second stage only adds about 1% to the total
> noise.
>
>
> - Show quoted text -
Yup, Unless the second stage has a higher bandwidth than the first
stage. (which I think might be a good idea), I've been building amps
with 5 or 6 x10 gain stages (to get thermal noise up to the 10V level)
and I think the only reason that this hasn't 'bitten' me yet is that
my bandwidth has been either limited by the first stage, or the slew
rate of the final stage.
George H.
John Larkin
Are you building a noise generator?
An 8 or 10 volt zener, 1n751-series maybe, run at maybe 1 mA, makes
about 300 nv/rthz wideband noise, which is easier to amplify than
resistor noise. And it swamps most opamp noise, which can be very
non-ideal.
John
Electronworks.co.uk
"John Larkin" <jjla...@highNOTlandTHIStechnologyPART.com> wrote in message
news:mkq65558jnsimkb1b...@4ax.com...
Surely this is talking about noise referred back to the input, not the
actual noise itself....
Or do I need another coffee?
John Larkin
I'm talking about the total noise that you see at the output. Divide
by the overall gain to get input-referred noise. What usually matters
in an amplifier is the s/n ratio at the output.
John
George Herold
Nope. I want to measure kT and e by measuring noise.
I've done the Zener trick in the past. I found that I got the largest
noise out of the highest voltage zeners (not surprising I assume
higher voltage zeners have larger electron avalanches.) A single
zener has a bit of an asymmetrical voltage waveform. But there are
circuits using two zeners in series and then summing the current.
(I’ve never tried those.)
George H.
John Larkin
Oh, right, slipped my mind.
How about paralleling a bunch of BF862's? Or did I suggest that
already?
>
>I've done the Zener trick in the past. I found that I got the largest
>noise out of the highest voltage zeners (not surprising I assume
>higher voltage zeners have larger electron avalanches.) A single
>zener has a bit of an asymmetrical voltage waveform. But there are
>circuits using two zeners in series and then summing the current.
>(I�ve never tried those.)
Yup, I've noticed the small asymmetry. It gets progressively worse at
lower dc bias currents, and many (all?) zeners eventually break up
onto raggedy sawtooth oscillations.
Subtracting the noise from two zeners should cancel the asymmetry.
Better yet, use 2*N zeners.
http://en.wikipedia.org/wiki/Central_limit_theorem
John
George Herold
<jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
> On Wed, 8 Jul 2009 07:31:44 -0700 (PDT), George Herold
>
>
>
>
>
Yeah, I think subtracting them is right. The circuit looked like
this,
V(+)--RRR--+--K|--+--K|--+--RRR--V(-)
| | |
C GND C
C C
| |
+--RR--+--RR--+
|
OUT
where K| are zeners
George H.
George Herold
<jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
> On Wed, 8 Jul 2009 07:31:44 -0700 (PDT), George Herold
>
>
>
>
>
Ahh, Thanks John I think that explains it. I was trying to maximize
the noise which happened with high voltage zeners at low bias current
(10-20 uA or so) If I ever build one again I'll try higher bias
currents. Of course if I ever build it again I would change some
other things too. Are you ever embarrassed by the circuits you
designed 5-10 years ago? (OK reset that to the beginning of your
circuit building career which for me started (in earnest) 9 years
ago.)
George H.
John Larkin
<gghe...@gmail.com> wrote:
>On Jul 8, 11:55锟絘m, John Larkin
><jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
>> On Wed, 8 Jul 2009 07:31:44 -0700 (PDT), George Herold
>>
>>
>>
>>
>>
>> <ggher...@gmail.com> wrote:
>> >On Jul 7, 11:34锟絧m, John Larkin
>> ><jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
>> >> On Tue, 7 Jul 2009 20:22:36 -0700 (PDT), George Herold
>>
>> >> <ggher...@gmail.com> wrote:
>> >> >On Jul 7, 11:32锟絘m, John Larkin
>> >> >stage. (which I think might be a good idea), 锟絀've been building amps
>> >> >with 5 or 6 x10 gain stages (to get thermal noise up to the 10V level)
>> >> >and I think the only reason that this hasn't 'bitten' me yet is that
>> >> >my bandwidth has been either limited by the first stage, or the slew
>> >> >rate of the final stage.
>>
>> >> >George H.
>>
>> >> Are you building a noise generator?
>>
>> >> An 8 or 10 volt zener, 1n751-series maybe, run at maybe 1 mA, makes
>> >> about 300 nv/rthz wideband noise, which is easier to amplify than
>> >> resistor noise. And it swamps most opamp noise, which can be very
>> >> non-ideal.
>>
>> >> John- Hide quoted text -
>>
>> >> - Show quoted text -
>>
>> >"> Are you building a noise generator?"
>> >Nope. I want to measure kT and e by measuring noise.
>>
>> Oh, right, slipped my mind.
>>
>> How about paralleling a bunch of BF862's? Or did I suggest that
>> already?
>>
>>
>>
>> >I've done the Zener trick in the past. 锟絀 found that I got the largest
>> >noise out of the highest voltage zeners (not surprising I assume
>> >higher voltage zeners have larger electron avalanches.) 锟紸 single
>> >zener has a bit of an asymmetrical 锟絭oltage waveform. 锟紹ut there are
>> >circuits using two zeners in series and then summing the current.
>> >(I锟絭e never tried those.)
>>
>> Yup, I've noticed the small asymmetry. It gets progressively worse at
>> lower dc bias currents, and many (all?) zeners eventually break up
>> onto raggedy sawtooth oscillations.
>>
>> Subtracting the noise from two zeners should cancel the asymmetry.
>> Better yet, use 2*N zeners.
>>
>> http://en.wikipedia.org/wiki/Central_limit_theorem
>>
>> John- Hide quoted text -
>>
>> - Show quoted text -
>
>"Yup, I've noticed the small asymmetry. It gets progressively worse at
>> lower dc bias currents,"
>
>Ahh, Thanks John I think that explains it. I was trying to maximize
>the noise which happened with high voltage zeners at low bias current
>(10-20 uA or so) If I ever build one again I'll try higher bias
>currents. Of course if I ever build it again I would change some
>other things too. Are you ever embarrassed by the circuits you
>designed 5-10 years ago? (OK reset that to the beginning of your
>circuit building career which for me started (in earnest) 9 years
>ago.)
>
>George H.
>
Well, I've been designing circuits since I gave up chemistry in about
1956 or so. Sure, I've done some terrible stuff, partly out of
ignorance and partly because parts used to be a lot worse and a lot
more expensive than they are now.
This is sort of a golden age now, with so many cool parts that can
implement all sorts of signal processing and algorithms that were only
theoretical not too long ago, or at least theoretical unless you had a
military-scale budget and a supercomputer.
John
JosephKK
While the noise output is highest in the knee, the asymmetry is also
highest there as well. Subtracting two of them is a great idea to
reduce the asymmetry some.
mi...@sushi.com
<jjSNIPlar...@highTHISlandtechnology.com> wrote:
> On Mon, 6 Jul 2009 10:40:04 -0700, "Bob Eld" <nsmontas...@yahoo.com>
> wrote:
>
>
>
> >"Bob.Jones5...@gmail.com" <bob.jones5...@gmail.com> wrote in message
> >news:ac7736e1-df1f-470e...@r33g2000yqn.googlegroups.com...
> >> Is there ever a time when it would be better to have multiple gain
> >> stages of a signal rather than use a single stage if the need is to
> >> use a single stage?
>
> >> If each stage has gain G_i why can't we always just choose G_1*...*G_2
> >> as the gain for the first stage assuming the op amp can handle it and
> >> we can accurately find the resistors?
>
> >Your question is not very clear. But, in general, each gain stage adds and
> >amplifies noise plus each stage has a finite bandwidth. Cascaded stages
> >reduce overall band width and increase noise Therfore you should use as few
> >stages as possible for a given situation.
>
> If the first stage has significant gain, the noise of downstream
> stages don't matter.
>
> And more opamps increase available gain-bandwidth, not decrease it.
>
> John
I think the other poster might want to brush up on input referred
noise. If you do it right, the first stage dominates.
baron
Just curious ! Have you tried an LED as a noise source ?
--
Best Regards:
Baron.
Michael A. Terrell
baron wrote:
>
> Just curious ! Have you tried an LED as a noise source ?
Or the classic noise generator used in W.W.II. A photo tube & lamp.
Different lamps had different characteristics, but some were capable of
wiping out the entire usable RF spectrum from LW to VHF and was used in
airborne jamming.
Fair Radio still had new surplus units, well into the early '70s.
--
You can't have a sense of humor, if you have no sense!
George Herold
> On Wed, 08 Jul 2009 08:55:13 -0700, John Larkin
>
>
>
>
>
> <jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
> >On Wed, 8 Jul 2009 07:31:44 -0700 (PDT), George Herold
>
> - Show quoted text -
JosephKK wrote,
" While the noise output is highest in the knee"
Joseph, do you have any idea why this is? I would naively expect that
the noise would only depend on the ‘avalanche factor’ (how many more
electrons are 'released' for each electron emitted into the
junction.) And I’d expect that to be mostly constant for a given
zener. So there must be something else going on.
George H.
George Herold
<jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
> On Wed, 8 Jul 2009 19:40:37 -0700 (PDT), George Herold
>
>
>
>
>
> >On Jul 8, 11:55 am, John Larkin
> ><jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
> >> On Wed, 8 Jul 2009 07:31:44 -0700 (PDT), George Herold
>
> >> <ggher...@gmail.com> wrote:
> >> ><jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
> >> >> On Tue, 7 Jul 2009 20:22:36 -0700 (PDT), George Herold
>
> >> >> <ggher...@gmail.com> wrote:
> >> >> >with 5 or 6 x10 gain stages (to get thermal noise up to the 10V level)
> >> >> >and I think the only reason that this hasn't 'bitten' me yet is that
> >> >> >my bandwidth has been either limited by the first stage, or the slew
> >> >> >rate of the final stage.
>
> >> >> >George H.
>
> >> >> Are you building a noise generator?
>
> >> >> An 8 or 10 volt zener, 1n751-series maybe, run at maybe 1 mA, makes
> >> >> about 300 nv/rthz wideband noise, which is easier to amplify than
> >> >> resistor noise. And it swamps most opamp noise, which can be very
> >> >> non-ideal.
>
> >> >> John- Hide quoted text -
>
> >> >> - Show quoted text -
>
> >> >"> Are you building a noise generator?"
> >> >Nope. I want to measure kT and e by measuring noise.
>
> >> Oh, right, slipped my mind.
>
> >> How about paralleling a bunch of BF862's? Or did I suggest that
> >> already?
>
> >> >noise out of the highest voltage zeners (not surprising I assume
> >> >zener has a bit of an asymmetrical voltage waveform. But there are
> >> >circuits using two zeners in series and then summing the current.
>
> - Show quoted text -
It's mostly my past ignorance that bothers me. (And of course I'm
still ignorant of so much...) Just looking back on the zener noise
circuit. Indeed I get about 300nV/rtHz of noise from the zener. (A
number I think you quoted earlier.) The 'mistake' I made in the
circit was using an opamp with not enough slew rate. At the highest
gain level the noise signal turns into a triangle wave...Sigh. Of
course I don't think anyone but me has noticed this and it hardly
matters.
George H.
George Herold
> George Herold Inscribed thus:
>
>
>
>
>
> > On Jul 7, 11:34 pm, John Larkin
> > <jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
> >> On Tue, 7 Jul 2009 20:22:36 -0700 (PDT), George Herold
>
> >> <ggher...@gmail.com> wrote:
> >> >On Jul 7, 11:32 am, John Larkin
> >> ><jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
> >> >> On Tue, 7 Jul 2009 07:40:39 +0100, "Electronworks.co.uk"
>
> >> >> <newsgro...@electronworks.co.uk> wrote:
>
> >> >> >"John Larkin" <jjSNIPlar...@highTHISlandtechnology.com> wrote in
>
> - Show quoted text -
Hi Baron, How are you thinking of using the LED? You can get full
shot noise in the current from any diode. I've also used LED's
shinning on photodiodes to get full shot noise. The advantage of an
avalanche zener diode is that you get ‘gain’ in the diode. Each
electron makes many more. You can also get a lot of shot noise from a
PMT used in current mode. There the gain is even bigger 10^6 or
so.
George H.
John Larkin
I was discussing ignorance with one of my guys yesterday. We're trying
to specify and buy a crystal oscillator to be used as the timebase for
an a/d conversion of a spin resonance effect. Spins may last seconds,
and multiple shots may be signal-averaged over up to an hour, and
we'll be doing FFTs on the averaged sample data. So how do we specify
the oscillator? We concluded that 1) we're not smart enough to really
understand the problem, especially in the way that OCXO makers specify
their parts and 2) if one of us did know this much about this or
similar subjects, we'd probably be too specialized to design the whole
system.
Can anybody recommend a source for OCXOs that doesn't want an insane
amount of money for them? 64 or 128 MHz would be nice, but we could
work with 5 or 10 or some such if that was a standard, affordable
part. Under $50 would work.
John
George Herold
What does the first "O" stand for? I had a job interview once at
Frequency Electronics. They make CXO's and atomic clocks. I had an
idea of how (maybe) to make a better temperature control. But I
learned that temperature drifts are not the big problem. The
'uncontrolled' frequency drifts come from material either leaving or
sticking to the Xtal. Can you synchronize the oscillator to the GPS?
We have ‘free’ atomic clocks orbiting over our heads.
George H.
John Larkin
"O is for "oven."
This particular app needs low phase noise and good stability for, say,
a hour at a time. Oven crystals run at their "turning point"
temperature (the flat spot on the f-vs-t curve) so are very stable
there. Longterm aging isn't an issue. Using GPS usually means locking
to a 1 PPS source, which is great for longterm stability but not much
help with phase noise. And GPS would sure blow the $50 budget.
John
Baron
Hi George,
The reason I asked was because I once built a device for optimising the
noise bandwidth of VHF receivers. Not being able to afford the
specified noise diode I tried a red led. I was quite surprised that it
produced as much noise as it did. It seemed to work quite well. I
still have that bit of kit kicking around.
Thankyou for your interesting notes.
--
Best Regards:
Baron.
George Herold
Baron, I was reading about zeners in AoE (2nd edition section 6.14)
during lunch today. And realized I'd been making a big mistake in my
understanding of the noise from zeners. I had been thinking it was
current noise. But it's voltage noise that is sensed. Ahhh, it’s
the I-V curve of the zener that turns the current noise into voltage
noise! As you get closer to the knee the curve flattens out and you
get a larger voltage noise for the same amount of current noise. (See
figure 6.22 in AoE) (Of course as you reduce the current you have
less current noise, but apparently the slope of the I-V curve is the
larger effect.) I think this also explains the asymmetry that is
observed.
So I think I now see how you used an LED as a noise source. I’ll have
to try it some time.
George H.
JosephKK
<gghe...@gmail.com> wrote:
>On Jul 9, 12:09 am, "JosephKK"<quiettechb...@yahoo.com> wrote:
>> On Wed, 08 Jul 2009 08:55:13 -0700, John Larkin
>>
<snip>
>> While the noise output is highest in the knee, the asymmetry is also
>> highest there as well. Subtracting two of them is a great idea to
>> reduce the asymmetry some.- Hide quoted text -
>>
>> - Show quoted text -
>
>JosephKK wrote,
>
>" While the noise output is highest in the knee"
>
>Joseph, do you have any idea why this is? I would naively expect that
>the noise would only depend on the ‘avalanche factor’ (how many more
>electrons are 'released' for each electron emitted into the
>junction.) And I’d expect that to be mostly constant for a given
>zener. So there must be something else going on.
>
>George H.
>
Quite simply it has to do with the curvature of the curve at the knee.
It is an unstable area in the operating characteristic, thus high
noise. However the changing slope perturbs (skews the third order and
fourth order moments of the probability distribution) the amount of
positive values compared to the negative values. But i would have to
study up quite a bit to back that up with real math.
George Herold
> On Thu, 9 Jul 2009 06:47:01 -0700 (PDT), George Herold
>
>
> - Show quoted text -
Thanks for the response Joseph, I 'think' I figured this out today.
You're right it has to do with the curvature of the I-V curve. All
diodes have shot noise in the current. (And zeners have more.) From
the I-V curve you can determine the conversion of current to voltage
and thus the amount of voltage noise. And the asymmetry is simply the
result of the curvature. A noise fluctuation that decreases the
current has a bigger effect on the voltage than one that increases the
current.
George H.
John Larkin
In the zeners I played with, as the current went down the noise began
to look asymmetric, then started to look more like random discrete
spikes, at at low currents turned into a noisy sawtooth oscillation.
That's not shot noise. It's more like random spots of negative
resistance. The voltage builds up into the junction capacitance, and
some small zone gets a thermal electron that starts a local avalanche
breakdown, which discharges the junction cap below the avalanche
voltage threshold. Something like that.
At high currents, there are many, many avalanche breakdown pulses
added up, so the central limit theorem kicks in to make the noise
gaussian.
The voltages involved are way bigger than shot noise would produce...
300 nV per root Hz with 1 mA into a few ohms of zener dynamic
impedance is roughly 3000 times the shot noise level.
John
George Herold
<jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
> On Thu, 9 Jul 2009 20:52:26 -0700 (PDT), George Herold
>
>
>
>
>
>
> - Show quoted text -
earlier. I’ll certainly look into it. To be honest I never
measured the I-V curve of a zener before yesterday. I used a 1n5245
15V zener and found about a 1 M ohm impedance up to about 14.2 Volts,
at which point it dropped to maybe 20 ohms or so. I had a crappy
meter measuring the voltage and not much resolution.
I (uA) V (V)
7 6.85
9.1 9.0
12 12.1
14.2 14
21 14.2
51 14.28
100 14.28
813 14.30
Sorry, I should have put more current through it but it was the end of
the day, and I had to get home.
The circuit I used to make a noise source uses a 20 V zener in
series with a 1Mohm resistor across the +/-15 Volt power rails. A
1uF cap picks off the noise voltage at the junction. So It looks like
I was running at a point ‘above’ the knee.
Call it 10uA of current which gives sqrt(2*e*I) ~ 1.8 pA/rtHz of
noise. Times 1 meg (I’m assuming the same resistance as above which
could be a mistake.) is 1.8 uV/rtHz. OK that is more noise than I
thought I had? Clearly more measurements are in order. But it’s
going to have to wait for a few weeks. I’ve got an important
conference/show in two weeks and I have lots to get ready.
then John wrote,
“> The voltages involved are way bigger than shot noise would
produce...
> 300 nV per root Hz with 1 mA into a few ohms of zener dynamic
> impedance is roughly 3000 times the shot noise level.”
>
Hmm, maybe we’re working in different ranges? I was working at 10uA
with a bandwidth of a few hundred kilo hertz max. (the 1uF cap was
followed by a 8 MHz GBW opamp with a gain of 20.)
And I wouldn’t be surprised that the current noise is way above the
‘single electron’ shot noise. As long as the time scales one is
looking at are long compared to the transit time of charges through
the diode, then the effective charge of the electron will be
multiplied by the Multiplication factor.
I noise = sqrt( M*e*I)
The only expression I found for M was the following.
M = 1/{1-(Va/Vbr)^n} where n is 2-6. (not very useful given the
range of n)
from here.
http://ecee.colorado.edu/~bart/book/book/chapter4/ch4_5.htm
This sounds like it will be fun to measure. First the current noise
then the voltage noise and then see if they fit with the DC I-V
curve.
George H.
Howard Swain
"John Larkin" <jjla...@highNOTlandTHIStechnologyPART.com> wrote in message news:08ie55dotu13rlavs...@4ax.com...
You might be interested in early work on avalanche diodes as noise
sources.
Haitz and Voltmer, "Noise of a Self-sustaining Avalanche Discharge
in Silicon: Studies at Microwave Frequencies," J. Appl. Phys.,
vol 39 (June 1968), pp. 3379-3384.
(In that paper they were running mA of current and getting noise at
Ghz frequencies that was, say, 20 dB above kT.)
Haitz, "Noise of a Self-sustaining Avalanche Discharge in Silicon:
:Low Frequency Noise Studies," J. Appl. Phys., Vol 38 (June 1967),
pp. 2935-2946.
I believe these draw on the theory of Hines in:
"Noise Theory for the Read Type Avalanche Diode," IEEE Trans.
of Electron Devices, vol. ED-13 (Jan. 1966), pp. 158-163.
Also, see this article by Motohisa Kanda of NBS:
"An Improved Solid-State Noise Source, IEEE Trans. on MTT-24
(Dec. 1976), pp. 990-995. (This covered 2 - 4 GHz)
--
Regards,
Howard
hsw...@ix.netcom.com
Howard Swain
You might want to go the Agilent website and look at the specs
for the 10811 oscillator to see the kind of things people spec.
(It appears they don't sell it anymore and I'm sure it was more than $50.)
I don't know enough about your need, but I wonder if the
Allan Variance might be how you need to spec it.
See D. W. Allan, "Statistics of Atomic Frequency Standards,"
Proc. IEEE, vol. 54 (Feb 1966), pp. 221-230.
Various people are now making "mini-rubidium" sources now,
I believe. Probably also more than $50, but I'd guess they
would be more stable over your time periods.
John Larkin
Stanford Research makes an equivalent, with an SC-cut crystal. We have
one, which we put into a rack-mount chassis with a power supply and
some buffers and dividers. The osc alone cost about $600.
>
>I don't know enough about your need, but I wonder if the
>Allan Variance might be how you need to spec it.
>See D. W. Allan, "Statistics of Atomic Frequency Standards,"
>Proc. IEEE, vol. 54 (Feb 1966), pp. 221-230.
>
>Various people are now making "mini-rubidium" sources now,
>I believe. Probably also more than $50, but I'd guess they
>would be more stable over your time periods.
>
I wonder how good the rubidiums and GPSs are for jitter in the seconds
sort of range. I'm guessing not much better than the basci
undisciplined XOs inside, since both the rubidium and the GPS are very
slow corrections.
The SRS has jitter of a few picoseconds per second of delay, about
100x better than I need, and about 10x our budget. And it's huge.
We can get an SC-cut crystal with a oven-compatible turning-point
temperature for around $20. But designing low-noise overtone crystal
oscillators is a nuisance.
John
JosephKK
And how many posters in this NG (s.e.d) do you think have regular
access to all these articles? For that matter any NG.
Anyone can point to any amount of papers, theses, articles or any
other publications not generally available and claim very much. This
is USENET. Do try to properly report the reasoning of these articles
as well as reference them.
George Herold
>
> - Show quoted text -
Oops. Please ignore the data in the above post. I realized driving
in today that at low currents I was just measuring the input impedance
of the voltmeter. (Idiot! why didn't the 1Meg ohm tip me off)
George H.
George Herold
> "John Larkin" <jjlar...@highNOTlandTHIStechnologyPART.com> wrote in messagenews:08ie55dotu13rlavs...@4ax.com...
> > On Thu, 9 Jul 2009 20:52:26 -0700 (PDT), George Herold
>
> - Show quoted text -
Excellent Howard I'll look up the references.
George H.
George Herold
> On Sun, 12 Jul 2009 17:08:44 -0700, "Howard Swain"
>
>
>
>
>
> <hsw...@ix.netcom.com> wrote:
>
> >> On Thu, 9 Jul 2009 20:52:26 -0700 (PDT), George Herold
>
> - Show quoted text -
I appreciate the references. I’m an alumina member of my college and
as such can go into the library and look up any references that they
have access to. I think it is the main reason that I give them money
every year.
George H.
Phil Hobbs
Rubidium and hydrogen are both better at short times than caesium. If
the volume would support the engineering required, how about using N
cell phone VCXOs and getting rid of the phase noise via closure? (N
oscillators give you (N**2-N)/2 phase measurements, which for N>2 is
overdetermined.)
Cheers
Phil Hobbs
--
Dr Philip C D Hobbs
Principal
ElectroOptical Innovations
55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058
hobbs at electrooptical dot net
http://electrooptical.net
John Larkin
<pcdhSpamM...@electrooptical.net> wrote:
>> I wonder how good the rubidiums and GPSs are for jitter in the seconds
>> sort of range. I'm guessing not much better than the basci
>> undisciplined XOs inside, since both the rubidium and the GPS are very
>> slow corrections.
>>
>> The SRS has jitter of a few picoseconds per second of delay, about
>> 100x better than I need, and about 10x our budget. And it's huge.
>>
>> We can get an SC-cut crystal with a oven-compatible turning-point
>> temperature for around $20. But designing low-noise overtone crystal
>> oscillators is a nuisance.
>>
>> John
>>
>
>Rubidium and hydrogen are both better at short times than caesium. If
>the volume would support the engineering required, how about using N
>cell phone VCXOs and getting rid of the phase noise via closure? (N
>oscillators give you (N**2-N)/2 phase measurements, which for N>2 is
>overdetermined.)
>
>Cheers
>
>Phil Hobbs
Think of this as a sonar-type thing. We ping the process with an
impulse and digitize the received signal. The frequency range is up to
20 MHz and the total echo time can be up to 10 seconds. The signals
can be very weak, so we time-average a lot of shots, over up to an
hour maybe. So in order that the shots dum nicely, we need the sample
clock to change less than a few ns per 10-second event, over an hour.
That's a few hundred PPB effective phase noise+drift.
We took two of our benchtop digital delay generators and triggered
them together. One was set to make a 10 second delay, the other 10
seconds+200ns. Then we used an HP 5370 time-interval counter to
measure the 200 ns difference and the standard deviation of same.
Using the standard, pretty-good TCXO versions of the DDGs, we got
around 38 ns RMS jitter on the 200 ns delta. Not good, considering
that a 20 MHz signal has a period of 50 ns. Signal averaging would be
wiped out.
We sent the units down to manufacturing and they upgraded them to
OCXOs; I've always wanted an OCXO in mine anyhow. That dropped the
jitter to about 2 ns RMS. Divide by root 2, and it looks pretty good.
People do a similar thing to measure oscillator phase noise: get two
identical units, tune them to slightly different frequencies, run them
into a mixer, and sic a spectrum analyser onto the resulting
difference signal.
So it looks like we need to buy an OCXO. OCXOs run the crystal at the
flat spot on its parabolic frequency/temperature curve. TCXOs probably
aren't as good because the thermal time constant of the crystal isn't
perfectly matched to the tau of the temperature sensor. Putting a
cover over a TCXO helps a lot by lowpass filtering millikelvin
temperature fluctuations.
John
Phil Hobbs
Yep. OCXOs are definitely better for long times, but they aren't
particularly cheap. You used to be able to get ovens for just the
crystal, which were a big help though not as good as ovenizing the whole
works.
GPS-disciplined oscillators are a bit like the AC grid--not great at
short times, really good at long times--which isn't a good match to the
requirement. Closure is one way to lock a group of oscillators together
such that the random parts of their variations are reduced--obviously
since there are only relative phase measurements, there's an overall
phase term exp(j omega t + phi) that you can't get rid of. Still, if
you have the FPGA space, it would be N times better, and would at least
meet the cost target.
Cheers,
John Larkin
<pcdhSpamM...@electrooptical.net> wrote:
I think N would be unreasonable for cheap VCXOs... ballpark 100.
Lap-Tech will make us an oven-specified SC-cut, glass case, vacuum
sealed overtone crystal for $20 or so. SC crystals used to cost over
$100, so I guess the processing has evolved a lot. So maybe I'll do my
own ovenized oscillator. We have one in production already, a
surface-mount board on a heated aluminum block, with the crystal
buried in the block. I might need a new circuit and board, but at
least we have the mechanics and thermals and test fixtures worked out.
Dang, I don't like designing overtone crystal oscillators.
The Lap-Tech crystals are beautiful.
ftp://jjlarkin.lmi.net/Lap-Tech.JPG
John
George Herold
>
> - Show quoted text -
"Think of this as a sonar-type thing. We ping the process with an
> impulse and digitize the received signal. The frequency range is up to
> 20 MHz and the total echo time can be up to 10 seconds."
Sounds like a free induction decay from an NMR. 10 seconds would mean
darn good magnetic field homogeneity or a really small sample.
George H.