Sciencemadness Discussion Board

Properties of avalanche noise in reverse biased PN-junction

woelen - 6-7-2016 at 06:17

At the moment I am experimenting with generation of avalanche noise in reverse biased semiconductor devices. This effect is quite interesting and can be used for interesting practical applications, such as filter analysis and testing of audio equipment.

I did some experiments with different types of PN-junctions and found quite different properties. Unfortunately I hardly can find information about my specific observations on internet. There of course is a lot of general information, the most important thing being written is that avalanche noise is white (which means that the power density is independent of frequency) and that there also can be mixed in other types of noise, most notably 1/f noise.

I did an experiment with an NPN transistor base-emitter junction, using different samples of the BC550C type. All of them produce a remarkable amount of avalanche noise, but the noise is far from white. It is close to gaussian (around a certain DC level). I used a circuit with 680K in series with the BC550C emitter-base junction and a power supply of 15 V. Parallel to the 680K resistor I have a fairly large capacitor (10 uF or so). I expected white noise from a few Hz to well beyond the audio band. The spectrum is flat from 1000 Hz to 50 kHz (this is the upper part of my home-built audio spectrum analyser), but below 1000 Hz it quickly falls off. At 500 Hz it already is several dB lower, at 100 Hz it is nearly 0. In my measurements I did not use the voltage across the base-emitter junction, but I used the current through it (I passed it through a simple linear amplifier with high transresistance). At first I suspected the DC cut-off frequency to be too low, but changing the capacitor value from e.g. 10 uF to 1 uF does not have any visible effect on the low frequency part of the noise spectrum. The cut-off frequency of the RC-combination is very much below the 500 Hz mark.
What could be the cause of the very low noise output for the lower part of the frequency range?

I also did an experiment with an 11 V zener. I connected this to a 15 V power supply, in series with a 750K resistor. Lower and higher values of this resistor lead to lower noise. It is interesting to see that this combination gives quite some noise, but a 10x amplification of the noise voltage is necessary for good measurement (now I did use noise voltage for measurement, blocking out the DC component). The noise is almost white, from 10 Hz to 50 kHz there is a small gradual decrease of appr. 1.5 dB. Below 10 Hz it drops down, but that is due to the blocking out of DC with a coupling capacitor. The small drop of 1.5 dB could be due to the presence of some 1/f noise in the noise as well.
The noise of the zener now has a nice frequency characteristic, but it is strongly asymmetric and is far from gaussian. In positive direction it has high narrow peaks, while in negative direction is has more rounded and wider valleys. After filtering out the DC component at a cut-off frequency of a few Hz (22 uF capacitor combined in series with 2.2K resistor at output of amplifier), the amplitude in positive direction on average at least is twice the amplitude in negative direction.
What could be the cause of this strong asymmetric behavior of the avalanche noise of a zener diode?

aga - 6-7-2016 at 08:09

This might be useful :-

https://books.google.es/books?id=uQlEr5iWOdUC&pg=PA176&a...

There is mention of the different operating modes (forward/avalanche, reverse/Zener) generating different noise characteristics.


Sulaiman - 6-7-2016 at 08:33

One thing to check,
frequency response of test setup with 680k to 750k source impedance,
(i.e. not directly from a relatively low source impedance sig-gen)
just to rule out unexpected responses.

EDIT: you must live far from civilisation ...
how are you 'seeing' what is going on vs. a.c. line + harmonics ?

EDIT: I guess that you may be measuring noise current via the 10uF IF your monitoring equipment is low-impedance.

Filtering out power supply noise is essential,
(e.g. 7 x 100k resistors in series with a bypass-noise-to-ground cap at each junction, no electrolytics)

One more; batteries make good, isolated, low-noise d.c. supplies

[Edited on 6-7-2016 by Sulaiman]

unionised - 6-7-2016 at 10:18

I'm confused. A diagram might help.
are you shunting the noise to the +ve power rail where the regulator will try very hard to remove it?


[Edited on 6-7-16 by unionised]



[Edited on 6-7-16 by unionised]

Noisy.jpg - 26kB

woelen - 6-7-2016 at 12:10

Just to make things clear:
- In the NPN-transistor experiment I connected a capacitor to the node at which the 680K resistor is and the emitter of the transistor. The other end of the capacitor is at GND. The base of the transistor is not connected to GND, but is used as input of a current to voltage amplifier with high transresistance. The amplifier has low impedance (which is good when its input is a current) and it has low output impedance (which is good when its output is a voltage). The output voltage is connected to a digital oscilloscope with FFT software. See following diagram:







circuit.png - 21kB

The resistor is equal to 680K, I tried different capacitors in the range 220 nF to 10 uF, all gave the same result. At the output of the amplifier I measure a noise signal with severely attenuated low frequency part (below 1 kHz), and flat up to 50 kHz.
In the next post I will show the other circuit.

wg48 - 6-7-2016 at 12:32

If your referring to the asymmetry in the wave form of the current thru the zener diode then: I think its principally amplified shot noise ie a random series of current pulses. If the dc component is filtered out with capacitive coupling then what you refer to as positive going and negative going are wrt the mean ac signal there is no reason why they should be equal unless the mean value happens to be equal to half the peak to peak value.

The current pulses may have a similar shape and width but the random interval between them may be much larger giving the asymmetry to the top and bottom of the waveform.

The slew rate of your amplifier can also produce asymmetry in a waveform.



woelen - 6-7-2016 at 12:40

The second circuit uses an 11 V zener diode as noise source. The high impedance voltage is connected to an emitter follower, which makes a low-impedance signal of this. The used transistor is a BC550C, which has a beta (current gain) of appr. 500 which is more than enough to drive the resistor of 22K. The voltage gain of the emitter follower is appr. 1 and the small signal amplitude was measured to be at most 30 mV, so this small delta of input voltage can easily be handled by the resistor of 22K (the total current through this must remain positive, but that always is the case, it is biased at well over 10V).

The output over the 10K resistor is connected to a high impedance voltage to voltage amplifier (not the same circuit as the one of the previous experiment).


circuit2.png - 58kB

I see no problems with my method of measuring the signals. The weak noise signals are well buffered and amplified and the measuring system in no way can influence the noise circuits.

I do not measure any 50 Hz hum in my signals. The test circuits are built compactly and I used small decoupling capacitors at the power supply rails to avoid picking up RF noise.

macckone - 6-7-2016 at 21:57

I did a study of the transistor noise circuit a long time ago.
My design was using a transistor amplifier.
And then matching impedance to a 50 ohm input.

One thing to note is that nothing is truly white noise as
maintaining the same peak to peak at progressively higher
frequency requires proportionally more power. So you eventually
run out of power. This is usually dealt with using a specified
cut-off. I used 50khz.

In my study, I used various compression and difference
algorithms to determine the actual entropy of the 'noise' with a
25khz sampling rate and a 16 bit sample. Only the lower 6 bits
were actually decent randomness. Which also says something
about the overall 'whiteness' of the noise.

woelen - 7-7-2016 at 01:54

Of course, true hypothetical white noise is completely random at any sample frequency, simply because the power density is a non-zero constant over the entire frequency range from DC to infinity. Real "white" noise is always frequency limited, and from a certain frequency (e.g. 50 kHz) it falls off. If you sample such a signal at a much higher rate than the cut-off frequency, then two adjacent samples will be close to each other. True hypothetical white noise does not exist, it would require infinite power. Practical white noise sources can go up to several GHz, but with simple home-made circuits and eBay-stuff you cannot achieve that.

My home-made oscilloscope (based on a XONAR U7 and a front-end, built around a pair of LME 49710 opamps) samples at 192 kHz and has a cut-off low pass filter designed to fall off quickly above 75 kHz. My FFT software only checks up to 50 kHz (much higher hardly makes sense with just 192 kHz sampling frequency and 75 kHz cut-off low-pass filter). So, in my system, adjacent samples certainly will be correlated and even some wave form can be reconstructed from it. Still, in the frequency range, studied in this experiment, one still can meaningfully say that something is white noise or not, based on the frequency spectrum of the signal.

So, using the two circuits I have now, either it is more or less white noise, or it is more or less gaussian, but not both. I want it to be both gaussian and white (within a certain frequency range of e.g. 10 Hz to 30 kHz).

macckone - 7-7-2016 at 07:37

I found significant sample correlation at 50% of the cutoff
frequency. To get truly white noise you are going to need to do a
lot of tweaking. One method I found satisfactory was to sample,
use that to generate random numbers then feed those into
a DAC at the required frequency. This produced good audio
frequency white noise.

aga - 7-7-2016 at 09:27

Quote: Originally posted by woelen  
So, using the two circuits I have now, either it is more or less white noise, or it is more or less gaussian, but not both.

Build both and mix the signals ?

wg48 - 7-7-2016 at 09:56

Quote: Originally posted by woelen  
Of course, true hypothetical white noise is completely random at any sample frequency, simply because the power density is a non-zero constant over the entire frequency range from DC to infinity. Real "white" noise is always frequency limited, and from a certain frequency (e.g. 50 kHz) it falls off. If you sample such a signal at a much higher rate than the cut-off frequency, then two adjacent samples will be close to each other. True hypothetical white noise does not exist, it would require infinite power. Practical white noise sources can go up to several GHz, but with simple home-made circuits and eBay-stuff you cannot achieve that.

My home-made oscilloscope (based on a XONAR U7 and a front-end, built around a pair of LME 49710 opamps) samples at 192 kHz and has a cut-off low pass filter designed to fall off quickly above 75 kHz. My FFT software only checks up to 50 kHz (much higher hardly makes sense with just 192 kHz sampling frequency and 75 kHz cut-off low-pass filter).

Snip

So, using the two circuits I have now, either it is more or less white noise, or it is more or less gaussian, but not both. I want it to be both gaussian and white (within a certain frequency range of e.g. 10 Hz to 30 kHz).


So you want a signal that has a frequency spectrum which is Gaussian profile and is flat (presumable to some spec) between 10Hz and 30kHz. Is that correct?