I did not expect this 0.33 uV as an absolute value of noise
across 3 LEDs giving a 5.67 Volt reference.
I would say nobody can calculate to get the last digit right.
Also in real circuit a current source or resistor will add some noise.
1uV noise across 1 Volt is equal to -120 dB
If we say 0.33uV is something like the noise, I can get across 5.67 Volt,
then we can express approximate noise level in decibel.
dB = 20 x log10(0.00000033/5.67)
Noise = -144 dB
that is low noise!
across 3 LEDs giving a 5.67 Volt reference.
I would say nobody can calculate to get the last digit right.
Also in real circuit a current source or resistor will add some noise.
1uV noise across 1 Volt is equal to -120 dB
If we say 0.33uV is something like the noise, I can get across 5.67 Volt,
then we can express approximate noise level in decibel.
dB = 20 x log10(0.00000033/5.67)
Noise = -144 dB
that is low noise!
Fascinating!
I notice a rough correlation in your data within types of increased noise with higher forward voltage, i.e., the reds with the highest noise had the highest Vf, same with the greens. I've noticed that different LEDs from different manufacturers (or types) show dramatically different dynamic impedance- I wonder how that might correlate?
My measurements indicate that the high efficiency and high output types all have higher dynamic impedance and higher Vf than older, crummier LEDs. But do they have higher noise, too...?
I notice a rough correlation in your data within types of increased noise with higher forward voltage, i.e., the reds with the highest noise had the highest Vf, same with the greens. I've noticed that different LEDs from different manufacturers (or types) show dramatically different dynamic impedance- I wonder how that might correlate?
My measurements indicate that the high efficiency and high output types all have higher dynamic impedance and higher Vf than older, crummier LEDs. But do they have higher noise, too...?
LEDs have a problem with darks spots and line defects, due to impurities in the crystal. The high noise level of LEDs at low currents may be due to the different forward voltage of these defects and the randomness of the current sharing between the defects and good LED material. This is the same problem as the zener vs avalanche breakdown choice making 6V zeners noisy.
Thanks for the enthusiasm guys, but don't over interpret these measurements. I think there are still a number of loose ends, and it would also be good if somebody could try to repeat at least some of the measurements, preferrably using both exactly the same component types I used as well as some other brands. I don't think this is yet to be conisdered facts of the type that should go in the Wiki, but feel free to put a link to this thread there if you feel that is appropriate. I think we'd better keep it in this thread for yet some time. What should be done is
1) That people scrutinize the methodology, test rig etc. that I used, in order to spot any serious errors.
2) Brainstorm about what further measurements should be done, or how these measurements could be repeated in an even better way. Forr already pointed out an important problem, that the noise contribution of the biasing CCSs can't be isolated from the measurements I did. That should be possible with some further measurements, I think.
I am also a bit surprised nobody ever asked for any details about how I did the actual RMS measurements from the signal output from the test rig.
As for the Wiki, it would be good if somebody who truly understands the noise issues could write at short tutorial, or link to soma appropriate articles. I am by no means an expert on this, having read only a few papers and also learnt some valueable things from John Curl and a few others. For instance, Johns is about the only one ever discussion shot noise, which seems to be the least understood and most poorly documented type of noise in semiconductors.
1) That people scrutinize the methodology, test rig etc. that I used, in order to spot any serious errors.
2) Brainstorm about what further measurements should be done, or how these measurements could be repeated in an even better way. Forr already pointed out an important problem, that the noise contribution of the biasing CCSs can't be isolated from the measurements I did. That should be possible with some further measurements, I think.
I am also a bit surprised nobody ever asked for any details about how I did the actual RMS measurements from the signal output from the test rig.
As for the Wiki, it would be good if somebody who truly understands the noise issues could write at short tutorial, or link to soma appropriate articles. I am by no means an expert on this, having read only a few papers and also learnt some valueable things from John Curl and a few others. For instance, Johns is about the only one ever discussion shot noise, which seems to be the least understood and most poorly documented type of noise in semiconductors.
As it happens, I made a noise measurement on BZX55C12V a few weeks ago. I wanted to compare it with the (85V) 85A2 neon reference, so I wired seven in series. The Zeners and the neon passed 5mA DC. Using a 22Hz-22kHz measurement bandwidth and a true-RMS rectifier, I measured noise at -67dBu (350uV) for the Zener string and -73dBu (170uV) for the neon. In terms of noise per volt of reference voltage, that's 4uV and 2uV respectively.
These are preliminary results.
These are preliminary results.
So if I've done my sums correctly, the LM329 produces 11uV over a 20kHz bandwidth, and as it's a 6.9V reference (just looked up the data sheet), that equates to 1.6uV per volt of reference voltage. That's 2dB quieter than the 85A2. OK, so I looked up the price. Just about 1 Euro at 10 off prices for the 100ppm version.
For this RED LED above 3-6 mA gives lowest noise figures.
I use 2-2.5 mA when using LEDs for reference voltages.
Like for current sources, or VREF in my discrete voltage regulators.
Just because I have seen this in audio circuits, 2-2.5 mA.
This is also a value of current I often use in small signal transistors.
A typical Voltage drop for some unknown brand of RED LEDs (2mm) I use
will be 1.55-1.60 Volt at 2.5 mA.
Christer found generally lowest noise values was at a current of 2-6 mA in LEDs.
The materials for red leds are the simplest to make and probably the easist to get pure and free of flaws. Blue leds are at the opposite extreme.
I think that bandgap devices are noisy because they work by amplifying the difference in voltage across two junctions at different currents. One of these is going to be very low and the tests this thread show low currents lead to high noise.
I think that bandgap devices are noisy because they work by amplifying the difference in voltage across two junctions at different currents. One of these is going to be very low and the tests this thread show low currents lead to high noise.
davidsrsb said:
If I've interpreted you correctly, an early red LED is likely to be quieter because the manufacturer was desperately struggling to make a revolutionary (but by modern standards, simple) device work at all, not tweaking known technology to screw more light out for a given current.
Interesting pojnt about bandgaps.
As for LEDs seemingly having an optimum bias current, there is a risk that the figures could be misleading. It could be that it is actually the CCS that has an increased noise at higher current leveles. Since I never tried to measure the noise of the CCSs separately, we cannot know this for sure. For those particular measurements I varied the emitter resistor in the CCS, but on the other hand, that means lower resistance for higher current, so that cannot be the reason. Still, we don't know what the results would be if using some other way of implementing the CCS.
As for bandwidths, it would be useful to measure the noise over certain selected bandwidths also, not just over the full audio band. It was suggested in another thread a while ago to at least use HP and LP filters at som low frequency like 100 Hz, to try getting an idea of the 1/f noise component. However, ideally one should try a number of filter frequencies like 10, 100 and 1000 Hz, I think. A very simple, but useful test could be to measure over the whole audio band but with a capacitor over the reference, since references are often used that way.
There is so much one could and should measure....
As for bandwidths, it would be useful to measure the noise over certain selected bandwidths also, not just over the full audio band. It was suggested in another thread a while ago to at least use HP and LP filters at som low frequency like 100 Hz, to try getting an idea of the 1/f noise component. However, ideally one should try a number of filter frequencies like 10, 100 and 1000 Hz, I think. A very simple, but useful test could be to measure over the whole audio band but with a capacitor over the reference, since references are often used that way.
There is so much one could and should measure....