mandat said:
Well, I only measured noise, so just like Jocko and someone
else said , you have to also consider the PSRR. If your design
alllows for a reasonably quiet supply to the reference , by
regulation or filtering for instance, and you use a CCS
then it may be reasonable to go by noise figures. If, on the other
hand, you use a simple circuit where you have a noise supply
and maybe use a resistor instead of a CCS then
PSRR may be a more important issue than noise. The
standard answer "it depends" always applies.
I studied noise as a separate issue, since that is inherent
in the components. The importance of good PSRR, ie. low
dynamic impedance, can to more or less extent be designed
away in a circuit. (Jocko, correct me if I'm wrong.)
Circlotron and others,
I don't think I will measure any laser diodes. There are many
components one might wish to test, but I think it is enough
for now. I treat this as a pilot
study which gives some partial answers and ask a number of
new and more specific questions that can be studied for those
who feel inclined to do so.
One should also remember that the TL431 is not primarily
designed to be a low-noise reference, but a precise
shunt regulator. According to my books a bandgap reference
has a well-defined voltage and is very temperature stable
(if designed right), and the TL431 seems to score well at
both these points.
designed to be a low-noise reference, but a precise
shunt regulator. According to my books a bandgap reference
has a well-defined voltage and is very temperature stable
(if designed right), and the TL431 seems to score well at
both these points.
Measurements for LED noise optimum
Since I started this thing I guess I'd better try to wrap it up, so
I have done some extra measurements on LEDs, using a single
current source with variable current. The purpose was to spot
any tendency towards an optimal current for lowest noise. The
differences in the 2-8mA range are so small that is mostly an
academic exercise to even bother about this, but since I am
an academic I was stupid enough to waste some time on it.
Obviously the sample set of types and devices is much to small
for any claims, but if one should nethertheless be bold enough
to try spotting some tendency in the measurements, there seems
most LEDs have an optimum somewhere in the 2 to 6 mA range.
Some prefer close to 2mA , some close to 6mA and some
somewhere inbetween.
Updated report attached. New measurements at the end
under heading Experiment 2.
I think I have had enough of measuring noise for now, so
don't expect any additional measurements. At least not
for voltage references.
Since I started this thing I guess I'd better try to wrap it up, so
I have done some extra measurements on LEDs, using a single
current source with variable current. The purpose was to spot
any tendency towards an optimal current for lowest noise. The
differences in the 2-8mA range are so small that is mostly an
academic exercise to even bother about this, but since I am
an academic I was stupid enough to waste some time on it.
Obviously the sample set of types and devices is much to small
for any claims, but if one should nethertheless be bold enough
to try spotting some tendency in the measurements, there seems
most LEDs have an optimum somewhere in the 2 to 6 mA range.
Some prefer close to 2mA , some close to 6mA and some
somewhere inbetween.
Updated report attached. New measurements at the end
under heading Experiment 2.
I think I have had enough of measuring noise for now, so
don't expect any additional measurements. At least not
for voltage references.
Cracking stuff!
Hello Christer, I'd like to thank you for taking the time and trouble to make all these measurements - they make fascinating reading.
I went through your v1.4 set of data, averaged across your five measurements for each condition, then normalised the noise compared to the forward drop to give a figure in uV/V. Looking at your LEDs, they typically fell out at around 0.16uV/V, with a very shallow noise null (only 1dB deep centred on 5mA +/-1mA). I compared across diode types, and taking the LEDs as the reference (all at 5mA unless specified):
12V 0.5W Avalanche: - 15dB
12V 1.3W Avalanche: -13dB
True Zener: +6dB
Forward biased silicon junction (1N4148 or BC549): +7dB
TL431 at 5V, with 1mA: +12dB
Transition between Zener and Avalanche: Much noisier, and very variable
TL431 at 2.5V (or 5V, with >1mA): +34dB
This is most surprising - your data indicates that the much-maligned 12V "Zener" diode is the quietest device by far.
Hello Christer, I'd like to thank you for taking the time and trouble to make all these measurements - they make fascinating reading.
I went through your v1.4 set of data, averaged across your five measurements for each condition, then normalised the noise compared to the forward drop to give a figure in uV/V. Looking at your LEDs, they typically fell out at around 0.16uV/V, with a very shallow noise null (only 1dB deep centred on 5mA +/-1mA). I compared across diode types, and taking the LEDs as the reference (all at 5mA unless specified):
12V 0.5W Avalanche: - 15dB
12V 1.3W Avalanche: -13dB
True Zener: +6dB
Forward biased silicon junction (1N4148 or BC549): +7dB
TL431 at 5V, with 1mA: +12dB
Transition between Zener and Avalanche: Much noisier, and very variable
TL431 at 2.5V (or 5V, with >1mA): +34dB
This is most surprising - your data indicates that the much-maligned 12V "Zener" diode is the quietest device by far.
Thanks Christer,
It sheds at least some light on the noise behaviour of several common devices used as a voltage reference. Next step would be noise versus frequency. 1/f noise starts at different places, at some devices at 100 Hz (or even lower), at others it starts at 1 kHz. Given the 20 kHz bandwidth of your measurements that doesn’t show up. Higher noise frequencies can be easily filtered away, but lower are more difficult. Anyway thanks for the great job done.
Cheers
It sheds at least some light on the noise behaviour of several common devices used as a voltage reference. Next step would be noise versus frequency. 1/f noise starts at different places, at some devices at 100 Hz (or even lower), at others it starts at 1 kHz. Given the 20 kHz bandwidth of your measurements that doesn’t show up. Higher noise frequencies can be easily filtered away, but lower are more difficult. Anyway thanks for the great job done.
Cheers
I initially intended to look at the spectrum too, but when i
relized how much time it took just to get the RMS
figures I didn't have enough patience left to do both. Not
for now at least. If anybody else volunteers to do that I
could mail the components I tested. With a few exceptions
I bought them specifically for this test and have kept them
separately to know which ones they are.
relized how much time it took just to get the RMS
figures I didn't have enough patience left to do both. Not
for now at least. If anybody else volunteers to do that I
could mail the components I tested. With a few exceptions
I bought them specifically for this test and have kept them
separately to know which ones they are.
Re: Re: What about impedance?
Anyone feel free to jump in and correct me if I'm wrong, but the dynamic resistance is used to describe the V/I relationship within the device and is not a true resistance and as such does not contribute to Johnson noise.
I suspect that the dynamic resistance will, however, affect the transfer of current noise into the voltage domain - introducing a noise voltage that is produced from the noise current from the current source.
Christer said:
Anyone feel free to jump in and correct me if I'm wrong, but the dynamic resistance is used to describe the V/I relationship within the device and is not a true resistance and as such does not contribute to Johnson noise.
I suspect that the dynamic resistance will, however, affect the transfer of current noise into the voltage domain - introducing a noise voltage that is produced from the noise current from the current source.
I'm not completely sure about this, but I think the noise contribution of the dynamic resistance is comparable to the noise contribution of the re of a BJT which is the shot noise of the emitter current with a value that is half of what would have been the thermal noise of an equivalent real resistance.
Steven
Steven
Code:
Originally posted by [I]Christer[/I]
TEST METHOD
-----------
For each type of DUT, two devices (denoted #1 and #2 and presumably
from the same batch) were tested at the three test currents 1, 5 and
20 mA and the equivalent noise at the DUT was measured and calculated
as described above. For each combination of device and current, five
10-second measurements were made.
For reference, the voltage drop at each test current
was also measured for one device of each type.
MEASUREMENTS
------------
All values are RMS values
Idle noise:
---------------------------------------------------
Measured idle noise of amplifier with grounded input:
0.19 0.19 0.19 0.19 0.18 uV
(The theoretical max value was calculated to 0.16 uV for
20kHz bandwidth and 0.22 uB for 40 kHz bandwidth).
Measured idle noise of amplifier with 100 Ohm source resistor:
0.26 0.25 0.24 0.26 0.26 uV
(The theoretical max value was calculated to 0.20 uV for
20kHz bandwidth and 0.28 uB for 40 kHz bandwidth).
Diodes:
----------------------------------------------------
1N4148:
#1 @ 1mA: 0.28 0.28 0.28 0.27 0.27 uV
#1 @ 5mA: 0.25 0.25 0.22 0.22 0.22 uV
#1 @ 20mA: 0.24 0.21 0.22 0.25 0.23 uV
#2 @ 1mA: 0.38 0.27 0.28 0.26 0.29 uV (Vf = 0.57 V)
#2 @ 5mA: 0.23 0.22 0.23 0.23 0.23 uV (Vf = 0.65 V)
#2 @ 20mA: 0.23 0.21 0.21 0.21 0.24 uV (Vf = 0.75 V)
LEDs:
Brand is Everlight unless otherwise stated.
----------------------------------------------------
EL202HD (red):
#1 @ 1mA: 0.31 0.32 0.31 0.31 0.32 uV
#1 @ 5mA: 0.26 0.26 0.27 0.27 0.27 uV
#1 @ 20mA: 0.39 0.36 0.37 0.36 0.37 uV
#2 @ 1mA: 0.39 0.37 0.38 0.38 0.35 uV (Vf = 1.82 V)
#2 @ 5mA: 0.32 0.30 0.30 0.30 0.31 uV (Vf = 1.89 V)
#2 @ 20mA: 0.41 0.40 0.41 0.41 0.46 uV (Vf = 2.09 V)
EL204GD (green):
#1 @ 1mA: 0.68 0.50 0.50 0.47 0.46 uV
#1 @ 5mA: 0.35 0.30 0.28 0.28 0.29 uV
#1 @ 20mA: 0.36 0.35 0.35 0.35 0.35 uV
#2 @ 1mA: 0.46 0.46 0.44 0.44 0.41 uV (Vf = 1.82 V)
#2 @ 5mA: 0.36 0.33 0.32 0.33 0.32 uV (Vf = 1.92 V)
#2 @ 20mA: 0.39 0.40 0.39 0.41 0.40 uV (Vf = 2.12 V)
Zeners:
All zeners of brand Temic.
----------------------------------------------------
BZX55/C5V6 (0.5W 5.6V):
#1 @ 1mA: 5.3 5.3 5.3 5.3 5.3 uV
#1 @ 5mA: 2.9 2.9 2.9 2.9 2.9 uV
#1 @ 20mA: 1.7 1.6 1.6 1.6 1.6 uV
#2 @ 1mA: 5.3 5.3 5.3 5.3 5.3 uV (Vr = 5.68 V)
#2 @ 5mA: 2.9 2.9 2.9 2.9 2.9 uV (Vr = 5.77 V)
#2 @ 20mA: 1.8 1.6 1.6 1.6 1.6 uV (Vr = 5.81 V)
BZX55/C12 (0.5W 12V):
#1 @ 1mA: 0.35 0.37 0.37 0.39 0.39 uV
#1 @ 5mA: 0.30 0.28 0.28 0.28 0.30 uV
#1 @ 20mA: 0.24 0.25 0.25 0.26 0.25 uV
#2 @ 1mA: 0.32 0.33 0.32 0.33 0.32 uV (Vr = 11.32 V)
#2 @ 5mA: 0.26 0.26 0.27 0.32 0.26 uV (Vr = 11.37 V)
#2 @ 20mA: 0.25 0.26 0.28 0.24 0.30 uV (vr = 11.42 V)
Miscellaneous:
-----------------
TL431 (Ref tied to cathode to work as 2.5V zener diode):
#1 @ 1mA: 20 20 20 20 20 uV
#1 @ 5mA: 20 20 20 20 20 uV
#1 @ 20mA: 20 20 20 20 20 uV
#2 @ 1mA: 20 20 20 20 20 uV
#2 @ 5mA: 20 20 20 20 20 uV
#2 @ 20mA: 20 20 20 20 20 uV
(No cheating, the measurements actually were so consistent.)
BC549 BE diode forward biased transistor:
#1 @ 1mA: 0.24 0.24 0.24 0.24 0.24 uV
#1 @ 5mA: 0.22 0.23 0.22 0.24 0.23 uV
#1 @ 20mA: 0.22 0.21 0.20 0.22 0.20 uV
#2 @ 1mA: 0.23 0.23 0.25 0.24 0.24 uV (Vf = 0.73 V)
#2 @ 5mA: 0.23 0.23 0.23 0.23 0.22 uV (Vf = 0.78 V)
#2 @ 20mA: 0.21 0.22 0.21 0.20 0.21 uV (Vf = 0.86 V)Download: Noise Measurements 1.4 by Christer
This topic deserves to be STICKY in forum.
It is very useful for designing low noise amplifiers, voltage regulators and other audio circuits.
Say I want a low noise reference at something like 5-6 Volt.
I can use one 5V6 Zener, TL431 Vref x2, 3 LEDs in series or BC549 Vbe x 8.
What am interested is a coefficient noise uV per Volt-reference. ( uV/V )
How can I get this from your figures, Christer ?
How do I strip your numbers from your test setup noise ( 0.20uV ) ?
How can I estimate resulting noise in my reference ?
Take Red LED at 5 mA
Measures: 0.27 uV and LED Voltage is 1.89 Volt.
3 LEDs in series will give 5.67 Volt
But what is the noise going to be like?
A - 3 x 0.27 = 0.81 uV
B - 3 x ( 0.27-0.20 ) uV = 0.21 uV
C - ( 3 x 0.27 ) - 0.20 = 0.61 uV
thanks!
Hi Lineup.
Thanks for finding my measurements useful. First note that my goal was not primarily to determine absolute noise figures for the components, but to find relative figures, i.e. how they performed relative to each other. However, if taking the measured idle noise of the test rig into account, the measured value for the 100 Ohm resistor correlates well with theory, so I suppose the other figures are reliable too. Of course this is still just a small sample set of brands and types etc. so be careful trusting generalizations.
Noise adds in RMS fashion. That is, if we have two noise voltages u and v, the total noise is sqrt(u^2+v^2). So let r be the idle noise of the test rig, m the meauserd noise and x the noise voltage of the DUT, then we have
m = sqrt(r^2 + x^2)
rearranging this gives
x = sqrt(m^2 - r^2)
For the 100 Ohm resistor I measured 0.26 uV (in most cases) and the idle noise was 0.19 uV (in most cases). Then the noise contribution of the resistor is
sqrt(0.26^2 - 0.19^2) = 0.18 uV
Unless my memory fails me, the Johnson noise of a resistance is
n = sqrt(4kTBR)
where k is Botzmanns constant, T is the temperature in K, B is the bandwidth and R the resistance. For T= 300, B = 20 kHz and R = 100 Ohm, we get 0.18 uV !!!!! Isn't it lovely to see how theory and measuremants actually can agree sometimes?
So the answer to your question, of how to get the noise voltage of the DUT alone is to use the rearranged formula above, just as I now did for the resistor. So in the case of the red LED that measures 0.27 uV, we get that the noise voltage of the LED itself ought to be
sqrt(0.27^2 - 0.19^2) = 0.19 uV
A chain of three LEDs should thus give a total noise voltage of
n = sqrt(0.19^2 + 0.19^2 + 0.19^2) = 0.33 uV
I never thought of testing this to further verify the measurements (stupid of me, but one can't think of everything) but theory then predicts that I should have measured a noise voltage of
n = sqrt(0.33^2 + 0.19^) = 0.38 uV
where 0.19 uV is the noise of the test rig.
So the answer to you second question is: neither A, B or C, but case D, the noise voltage is 0.33 uV for three LEDs.
Thanks for finding my measurements useful. First note that my goal was not primarily to determine absolute noise figures for the components, but to find relative figures, i.e. how they performed relative to each other. However, if taking the measured idle noise of the test rig into account, the measured value for the 100 Ohm resistor correlates well with theory, so I suppose the other figures are reliable too. Of course this is still just a small sample set of brands and types etc. so be careful trusting generalizations.
Noise adds in RMS fashion. That is, if we have two noise voltages u and v, the total noise is sqrt(u^2+v^2). So let r be the idle noise of the test rig, m the meauserd noise and x the noise voltage of the DUT, then we have
m = sqrt(r^2 + x^2)
rearranging this gives
x = sqrt(m^2 - r^2)
For the 100 Ohm resistor I measured 0.26 uV (in most cases) and the idle noise was 0.19 uV (in most cases). Then the noise contribution of the resistor is
sqrt(0.26^2 - 0.19^2) = 0.18 uV
Unless my memory fails me, the Johnson noise of a resistance is
n = sqrt(4kTBR)
where k is Botzmanns constant, T is the temperature in K, B is the bandwidth and R the resistance. For T= 300, B = 20 kHz and R = 100 Ohm, we get 0.18 uV !!!!! Isn't it lovely to see how theory and measuremants actually can agree sometimes?
So the answer to your question, of how to get the noise voltage of the DUT alone is to use the rearranged formula above, just as I now did for the resistor. So in the case of the red LED that measures 0.27 uV, we get that the noise voltage of the LED itself ought to be
sqrt(0.27^2 - 0.19^2) = 0.19 uV
A chain of three LEDs should thus give a total noise voltage of
n = sqrt(0.19^2 + 0.19^2 + 0.19^2) = 0.33 uV
I never thought of testing this to further verify the measurements (stupid of me, but one can't think of everything) but theory then predicts that I should have measured a noise voltage of
n = sqrt(0.33^2 + 0.19^) = 0.38 uV
where 0.19 uV is the noise of the test rig.
So the answer to you second question is: neither A, B or C, but case D
, the noise voltage is 0.33 uV for three LEDs.
Christer said:Hi Lineup.
Thanks for finding my measurements useful.
..............
So the answer to your question, of how to get the noise voltage of the DUT alone is to use the rearranged formula above, just as I now did for the resistor. So in the case of the red LED that measures 0.27 uV, we get that the noise voltage of the LED itself ought to be
sqrt(0.27^2 - 0.19^2) = 0.19 uV
A chain of three LEDs should thus give a total noise voltage of
n = sqrt(0.19^2 + 0.19^2 + 0.19^2) = 0.33 uV
I never thought of testing this to further verify the measurements (stupid of me, but one can't think of everything) but theory then predicts that I should have measured a noise voltage of
n = sqrt(0.33^2 + 0.19^) = 0.38 uV
where 0.19 uV is the noise of the test rig.
So the answer to you second question is: neither A, B or C, but case D
thanks very much, Christer
By the way this is the noise test rig circuit, version 1a, Christer used,
in case anybody wants to make further tests.
All these results answer to questions I've asked for long. Still then, the subject of noise of biasing sources was very poorly documented. Everybody would like more and more data, but these ones are the most wellcome and are already very useful to any designer. Thanks a lot, Christer.
~~~~~ Forr
§§§
~~~~~ Forr
§§§
forr said:
You are quite right. The noise of the current source(s) isn't included in the idle noise, so my previous answer to Lineup is not quite right. It's already long ago since I did those measurements, so I have som difficulties to remember exactly what I did. What I should have done, but obviously didn't do, is to also measure the noise of the resistor when biased with the three current sources. Then it would have been possible to calculate the noise of the DUT alone. As of now, the figure one gets from the calculations I explained earlier is the noise of the DUT and the bias CCS. Of course that makes the calculation of the noise from the three LEDs in series even more wrong. I am afraid we can't really produce any useful absolute figures then withouth further measurements. I obviously should have planned the experiments even more carefully, but I had never imagined I would be able to get such good measurements that they would be useful for anything but a coarse comparative study.
Well, I've said it before, and Lineup said it again, the test rig schematic is there for anybody to repeat the experiments.
(OK, maybe one day I'll do a part II on this, but it is not a high priority project for me.)