You're stuck with the wideband noise of the reference, though. I often have to care about nanovolt 1-Hz noise on power supplies.
Cheers
Phil Hobbs
-- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 hobbs at electrooptical dot net http://electrooptical.net
Yup. A two-pole cap multiplier (the above circuit with another RC stage in front of the regulator) can do 100-dB of suppression at SMPS frequencies, which is what I usually want it for.
Connecting the cold end of a photodiode to a supply that isn't super-quiet is one good way to blow the signal out of the water.
Cheers
Phil Hobbs
-- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 hobbs at electrooptical dot net http://electrooptical.net
That only helps within the regulator's bandwidth, and the maximum suppression is limited to its loop gain.
The cap multiplier is quiet at all frequencies of interest (usually), and the regulator just keeps its output from wandering around.
For running op amps, a barefoot cap multiplier is usually better, because it's simpler and quieter, very stiff at high frequency, and you usually don't care about a few hundred millivolts' worth of supply sag.
Cheers
Phil Hobbs
-- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 hobbs at electrooptical dot net http://electrooptical.net
[ AC-coupled example, with some DC stability issues unaddressed]
Classic solution: a string of mercury batteries. You can still do it, with standard cells.
Wideband noise in a '431 is likely recombination noise in the base currents, as I understand it. So, maybe one with higher quiescent current would be better. A FET solution (like current-limit diode for a reference and jFET op amp for gain) would help, if things like popcorn/flicker didn't take over.
Even the humble follower transistor in the lightly-loaded C-multiplier, might be noisier than a FET (or MOSFET,though I recall a discussion that showed the MOSFET to have no advantage at Zgate under 100k ohms).
There's one voltage source that has very low ripple, good filtering, and the same noise output as a low-value resistor. That's a thermopile. Maybe the heatpump items aren't optimized for it, but one could imagine using two-alloy printed wiring to get to some serious potential differences in a compact, rugged, ultrareliable component.
You'd just have to allow a bit of warm-up time on that photon detector.
Cap multipliers are totally cool, when it comes to low noise. If you haven't built one and tried it, you should. I needed a low noise ~10 Watt supply. (30V, 0.3A) I tried a couple of TIP31/ 32's. a few (1-3 nV/rtHz.)* (measured in 100Hz-100kHz band width) Done. George (don't argue with success) H.
*measured with an 8nV fet opamp
Horrible tempco, short lifetime, expensive, and completely unnecessary. Some circuits require very low DC drift and noise in their power supplies, but not many. Most of the time the low-frequency PSR of op amps and current sources and such like will take care of the low baseband instability.
The PSR of op amps falls off pretty badly with frequency, however, and the noise current due to photodiode capacitance differentiating its bias supply rises linearly. So you really want the supplies to be quiet at high frequency, plus you want excellent rejection of SMPS ripple. Cap multipliers are great for that, but normal voltage regulators aren't. Thus the shunt-regulated cap multiplier is a big win.
Bandgaps are inherently noisy, because they have to put 20 dB of gain on the delta-Vbe to get it to cancel out d(Vbe)/dT. Base current shot noise (which is what I think you're talking about) is rarely dominant at low current densities.
Your average low-sat transistor such as a ZXT11N15DFTA has about a nanovolt 1-Hz noise at 1 mA of collector current. Your average bandgap starts at 40 nV and goes up from there--a difference of 30 dB or more.
MOSFETs make horrible cap multipliers, because they're far noisier than BJTs, their operating V_GS is much bigger than a V_BE drop, and they have very low transconductance.
Ultrareliable isn't how I'd describe it. Also its tempco would be on the order of 3000 ppm/K.
Cheers
Phil Hobbs
-- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 hobbs at electrooptical dot net http://electrooptical.net
Not so. That's the tempco of the series resistance, not of the voltage output! You'd be differentially heating two (sets of) thermocouples, and a common-mode temperature doesn't change the voltage at all.
A thermopile is a bunch of thermocouples in series. Their output voltage is approximately proportional to the temperature drop across the junction pairs. That means that V = K(T_hot -T_cold).
Even if T_cold = 0 kelvin, that scheme can't do better than PTAT (3000 ppm/K at room temperature), and with feasible T_cold values, it's much worse.
Cheers
Phil Hobbs
-- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 hobbs at electrooptical dot net http://electrooptical.net
That is a tremendously improbable case! Thermopiles operated on HEAT SOURCES stabilize to a fixed gradient (established by heat conduction) and one expects (T_hot - T_cold) to be a constant. It is NEVER proportional to absolute temperature in normal circumstances.
It's true, by the third law of thermodynamics, that thermocouple coefficients (and other things) do vanish at absolute zero, but at room temperature a thermocouple has no capability to give an absolute temperature reading, and that's because the Seebeck equation contains no absolute temperature sensitivity.
You miss my point. The output voltage is proportional to the temperature drop across the junction, and so the tempco goes as
1/(Thot-Tcold), which is huge--generally _much bigger_ than PTAT. (Tcold being zero kelvin was a fictitious best case for your argument, with zero physicsy stuff in it--see the linear approximation above.)Thus there's absolutely no reason to go to all that trouble, when a simple shunt regulated cap multiplier will do much better for under a buck, with a 1-nanovolt noise floor.
Cheers
Phil Hobbs
-- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 hobbs at electrooptical dot net http://electrooptical.net
Which is why you need a cold junction reference to measure against.
Jamie
Actually, you need the physicsy stuff to find a regime where PTAT is the case. Take 100x100mm of flexible circuit board, put quarter-millimeter wires with quarter-millimeter spacings all over it. That's 200 thermocouples. Roll it into cigarette-size, glue a base on one end, and a heater resistor on the other. Insulate everything but the base.
Send enough current through the resistor to raise the hot end temperature
100 degrees above the ambient-temperature end (i.e. set the power at the thermal conductance of the wires times 100 degrees).Thot - Tambient = heat_input / thermal_conductance = 100
Now look at dependence on ambient temperature Ta
dV/dTa = d/dTa (K) * (Thot - Ta) + K * d/dTa ( Thot - Ta)
but we know that the sum Thot - Ta is set by the heat input, that sum is constant!
So the temperature dependence is PTAT whenever the thermocouple has a Seebeck coefficient that is proportional to temperature. That happens at/near absolute zero (third law of thermodynamics).
The principle is not at all impractical; accurate RF current meters have worked this way for decades, inside the box where one rarely sees it happening.
You're just blowing smoke. A very garden-variety voltage reference can do 30 ppm/K, and good ones are below 5 ppm/K (e.g. the LT1021). Good luck getting anywhere near that with a thermopile, especially one in SOT-23. ;)
Cheers
Phil Hobbs
-- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 hobbs at electrooptical dot net http://electrooptical.net
Ah, if only life were that simple. To get best errors from a biased photodiode, you need the bias voltage NOT to be temperature independent (because the diode capacitance, and stored charge, isn't). There's a square root of T in the formula...
What formula are you talking about, exactly?
The PD bias is there to reduce its capacitance (by as much as 7 times), and therefore to reduce the high frequency noise of the TIA, which is
i_N = e_NTIA * 2 pi f C_diode .
As long as the PD bias doesn't have significant noise in the bandwidth of interest, you don't care about small variations.
At one point, we were talking about voltage regulators, and the noise rejection thereof.
Cheers
Phil Hobbs
-- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 hobbs at electrooptical dot net http://electrooptical.net
My wife gave me this as a birthday present. Can't imagine where she got it.
-- John Larkin Highland Technology Inc www.highlandtechnology.com jlarkin at highlandtechnology dot com Precision electronic instrumentation
The discussion started with stabilizing a bias voltage for a photodiode; the current from a photodiode is
I = Iphoto + Isat + d/dt(C * Vbias)
one usually ignores (or calibrates out) the constant Isat That last term isn't just one error, it's the sum of two terms dC/dt * Vbias + C * d/dt(Vbias)
Thus, in a temperature-varying-with-time case, change 't' time for 'T', temperature, and optimum photocurrent measurement requires
0 = dC/dT * Vbias + C * d/dT(Vbias)d/dT( Vbias) = -( dC/dT * Vbias) / C
It's not worth worrying about unless low-frequency response is important, or unless the amplifier could saturate, because temperature is a slow-changing variable.
Sze (_Physics of Semiconductor Devices_) gives a formula for abrupt-junction diode capacitance that seems appropriate
C = [constant] * sqrt(Vbi - V - 2kT/q)
where Vbi is the builtin voltage, V is the bias...
There's no error term in the external circuit from dC/dt. It's caused by the spreading of the shielding regions on the edges of the depletion region, and doesn't have to be charged up from zero volts.
Have you ever actually designed a photoreceiver? Doesn't sound like it. You're just blowing smoke.
The temperature of the photodiode changes on a timescale of minutes to hours. If you have a capacitance of 10 pF at 25 C, it might change by a picofarad or so between there and 100 C. Typical parameters might be a
10V reverse bias, and a very generous estimate of the thermal TC for the assembly might be 30 seconds.For an extreme example, say you dunked the whole thing into boiling oil at 200 C, i.e. a 175K transient.
The current error resulting from your mechanism would be
delta I = (dT/dt)*dC/dT*V_bias
= (175 K)/(30s) * (1 pF/100 K) * 10 V = 0.6 pA.
Not very impressive numbers, even for an extreme case, and of course it would almost all be well below 1 Hz.
How many sensitive photoreceivers have you actually designed?
Cheers
Phil Hobbs
-- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 hobbs at electrooptical dot nethttp://electrooptical.net
I haven't seen one of those around in years.
Jamie
Of course there is! The charge on the capacitor doesn't move through the depletion region, it goes in the external wiring.
One, in particular, was for fluorescence spectroscopy on microscopic samples. It had a ~1 cm photodiode to cover the monochromator's exit slit, and took a minute or so to scan the spectrum with a stepping motor moving the grating. The output was picoamps. So was leakage. Capacitance was high, seconds-to-minutes response was what worked best with a chart recorder.
It mismatches by a few orders of magnitude the numbers you've plugged in.
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.