White noise produced by reverse NPN Emitter-Base voltage?

Several years ago I made an accidental discovery where connecting a transistor like is shown below in the schematic would produce an audible white noise.
This was before I had internet, and lived out in the middle of nowhere, so I had no idea how to explain the behavior, nor any sources that mentioned it.

I know that typically white noise is undesirable, but this method produced it with enough amplitude to be audible on a magnetic speaker, much more so with a crystal earphone, and easily amplified.
This seemed useful to me, it would allow for the creation of more natural sound effects without the need of a random number generator.


So why does it do this?
Why don't any of the simulation programs model this behavior? I know it strays from the primary purpose of a transistor, but it does have various applications none the less.
This has been bugging me for years, and my best guess was that it was some sort of effect caused by discharges across the depletion zone. (Which confuses me because the collector must also receive charges by a sort of 'static' means in normal operation, but no such noise is generated, it seems to be drowned out if there is any.)

This effect only occurred when I went up to about 9 volts, and used a resistor above 200k ohms, the best amplitude reached at around 360k ohms.

The noise is produced because you are reverse biasing the EB junction. It then behaves as a zener diode of about 5v.
Every zener diode produces noise when you "starve" it. That is, when it is operating near the knee region of its characteristic curve. At less than the recommended current.
There is a distinction between low voltage zeners ( < 6v) and the ones that operate at higher voltages that are avalanche diodes and their noise generation is different.
This effect is used in noise generators since long time. And there are diodes optimized to generate noise for this application. As this ones:
https://noisecom.com/products/components/nc100-200-300-400-series-chips-and-diodes
Calibrated noise generators are very useful in measuring RF devices and others.
I have seen noise generated by a zener and then converted to random TTL pulses, used to simulate stochastic behavior. About 35 years ago.

Avalanche noise is discussed in profound semiconductor and analog design text books. It's basically present in all zener diodes with dominant avalanche breakdown, as said. In simple terms, the shot noise common to all junction currents is multiplied by the avalanche effect. Each charge carrier pair freed in the junction causes a large current pulse.

I have been mostly using small signal transistors (e.g. BC107) in BE reverse breakdown for analog noise generators. I never thought about it in detail, but I guess, the small junction capacitance compared to regular Z-diodes results in a higher bandwidth.

Spice noise analysis only considers regular junction current generated shot and flicker noise, but no avalanche noise.

I noticed that a lot of the white noise generators out there use pre-recorded noise, or a number generator.
Why don't they use this method? It seems it would be cheaper since all one needs is the noise source, an amplifier, and a filter to produce a sound that isn't so harsh.

Many of those noise generators have issues with repetition, and the brain seems to be very well-suited for recognizing patterns if it doesn't go on for longer than a minute and has a flawless loop transition.

I noticed that a lot of the white noise generators out there use pre-recorded noise, or a number generator.

Click to expand...

Which noise generators do you refer to? Classical analog systems, e.g. the Moog Synthesiszer have been using analog noise generators. It's however not surprizing that digital laboratory signal generators or purely digital music instruments are using digital (pseudo random) noise.

I agree that simple maximum length sequences from linear feedback shift registers have clearly noticeable periodicity. I presume that increasing the sequence length and slight modifications of the sequence can hide audible periodicity effectively.

There's still a need for true random sequences in some applications that is served by analog noise generators.

P.S.: On the other hand, using known pseudo random signals makes sense for some applications as well, because the measurement uncertainty caused by the noise signal cancels out when averaging over full sequence periods, e.g. when measuring acoustic channels.

Hi friends;
By coincidence nowadays i am interesting similiar application, but i use zener diode cct in this link with some changes.
I sampled the signal and store on PC. Now i want to analyse if it is really random or not. Its distrubution resembles gaussain a bit. But i am not clear how to judge it. I want to measure it's closeness to gauss distrubution or extract some numeric metrics giving its randomness level etc.
Still searching as i have time. Do you know any method to achieve this kind of stuff?
Thank you.