Measuring Junction Noise

A Ciphers By Ritter Page

Terry Ritter

2006 February 13

Also see: Junction Noise Measurements I


Noise is inherent in analog electronics. Thermal noise is generated in every resistor whether connected to power or not. Shot noise is generated whenever current flows through a semiconductor junction, either in forward or reverse breakdown modes. But these noise sources are very, very weak. Detection requires a relatively good preamplifier, with shielding against our surprisingly interfering environment. Nevertheless, it is very possible to amplify and measure tiny amounts of produced noise. It is also possible to compare and contrast different noise sources, such as zener diodes, IC "zeners" and low-voltage Metal-Oxide-Varistors (MOV's).

Thermal Noise (Johnson noise) is a result of thermal agitation, the sum of the Brownian movement of electrons in a resistance, even when not in a circuit or when no current is applied. Thermal noise is the one-dimensional summation of the movement of vast numbers of individual electrons, each with different speed and direction. Thermal noise is a very weak noise source.

Shot Noise is a statistical effect of event "clumping" when the events have an expected rate but independent times. Shot noise does not occur in conductors, where electrons are correlated like water in a hose. Instead, shot noise requires some sort of independent emission of discrete events, such as electrons through a semiconductor junction or water drops through a spray nozzle. Shot noise is another very weak noise source.

1/f Noise (flicker noise) has various sources, but we can minimize the effect by rolling off low frequencies in a filter.

Zener Diodes are a common source of deliberate analog noise. A zener diode is little more than ordinary semiconductor junction with a known breakdown voltage. All diodes break down at some voltage, but most are intended to not break down in use. In contrast, zeners are intended to break down in normal operation. Thus, zeners are manufactured to break down at relatively low voltages, for use in semiconductor circuits.

Breakdown is almost never a single voltage, but is instead a curve which covers a range of voltages depending on the current. Manufacturing tolerances mean that devices of the same type may start to break down at slightly different voltages. Manufacturing differences mean that devices of the same type may break down differently.

Zener Breakdown occurs mainly in junctions which break down at low voltages, typically below 5V or so. True zener diodes typically have a fairly low noise level, much like raw shot noise.

Avalanche Breakdown dominates in junctions which break down above 5V, and often produces much more noise. Transistor base-emitter junctions generally break down from 6V to 8V or so, and are an example of avalanche breakdown.

Measurement Box Goals

Various issues are involved in developing the ability to hear and measure diode noise:

Shielding. The desired noise signals have microvolt levels. To protect them, at least electrostatic and electromagnetic shielding is required. I decided to use a "Danish cookie tin," which obviously is ferrous or magnetic, but the thin steel could not do much to keep out 60Hz magnetic fields.

Amplification. Microvolt level signals need to be amplified, inside the shielding, to a level which can better compete with the outside environment. I chose 30dB as a reasonable step. A total of 60dB of amplification is needed to reach the lower levels of my equipment which will measure signals of -60dBm. It is also important to listen to the noise, and to use speakers instead of headphones.

Current Adjust. A diode junction has some DC offset when biased with a current of particular level. Varying the current always varies the junction offset. (A zener of stated voltage only has that voltage in a circuit with a given current.) It is desired to vary the bias current over a wide range of magnitudes, here from zero and 1uA to 10mA.

Inactive Circuitry. Various active servo circuits could be introduced to simplify operation, but doing that might create some uncertainty about the actual source of the resulting noise. To avoid uncertainty, the circuit is inactive up to the point of first amplification.

External Measurement. The whole point of the exercise is to measure physical values, and we cannot look at a meter which is inside a metal box. Internal signals must be conveyed to the outside, bearing in mind that the whole point of the shielding is to keep outside signals from getting in.

Measurement Box Circuit

Basically, we have a potentiometer across battery power driving a resistor in series with a diode junction. A DC-blocking capacitor delivers junction voltage noise to a preamplifier, and then to the RMS measurement system.

The potentiometer is used to exceed the junction offset of the device under test (DUT). Five selectable precision resistors each produce a given current when 1.2V appears across them. We can measure that happy occurrence externally.

The potentiometer I used is a peculiar dual 3-turn precision wirewound in the old style. The switches are small toggles. The diode is held by tiny alligator clips on stiff wires. The diode battery, the first preamp stage, and the preamp battery all fit inside the shield with the top on.


Originally I wanted to measure the junction current directly, though test points A and B. Unfortunately, digital multimeters seem to add their own modulation to the flowing current, which can be sensed and heard at these very low levels. Accordingly, the current flow was jumpered internally. An old-style analog multimeter might be OK, although a wide range of currents do need to be sensed.

The inductor and capacitor (e.g., L1 and C1) at each test point A, B and C are inside a commercial lowpass filter or "feedthrough." These devices are fully shielded and are intended to keep outside RF outside. Unfortunately, it appears that some diodes will oscillate, and do so inside the shield, thus affecting the values delivered to the outside. Accordingly, R7, C4 and R8 were added to reduce measurement problems.

Measuring voltage across the test points B and C describes the amount of current through the junction. Measuring voltage across C and ground shows the bias voltage across the diode.

In practice, I have to remove the lid to change the diode. To set the current, I select a range on the switches, then adjust the voltage pot for 1.2V on meter M2. Then I replace the lid and measure the result.

The Preamp Signal

The DC-blocking capacitor C7 is large to support preamps which need a low-impedance signal source for lowest preamp noise.

Unlike resistive impedance, capacitive reactance does not create noise, but it can get in the way. When the device under test (DUT) has a particularly low dynamic impedance, noise from the preamp input device can be "shunted" into the DUT. That is, device noise flows out of the input of the preamp into the signal source, while signal from that source flows to the preamp as expected. Capacitor C7 must be large for the noise shunting action to be effective, even if the preamp has a high input impedance.

How large must C7 be? A preamp stage using an Analog Devices OP27 would like a source impedance under, say, 300 ohms. So the effect of C7 should be negligible if we keep the impedance under, say, 30 ohms. A 100uF capacitor will be 31.8 ohms capacitive at 50Hz, and correspondingly less at higher frequencies. At 10 seconds, the RC time constant is a little long, but we can live with it.

The ultimate advantage all this might deliver depends upon working with a junction or other device with a source dynamic impedance under 300 ohms. With ordinary semiconductor breakdown, that seems unlikely. However, an "IC zener" design might well have a dynamic impedance in the tens of ohms.

The Back-End System

After 30dB of amplification inside the measurement box, the audio noise signal exits on a double-shielded cable. It then enters a completely shielded preamp box, holding another 30dB gain stage, with its own battery power. Both preamp stages have a fixed, known gain so that measuring the amplitude of the resulting signal gives us the amplitude of the original signal.

Inside the preamp box, the signal first encounters a highpass RC filter stage targeted at 1000Hz. (In practice, 0.022uF and 7.5k should roll off below about 965Hz.) That filter takes out most 60Hz energy, which can occur strongly from time to time and cause signal clipping.

The highpass filter output goes into the 30dB gain stage, and then into another RC filter. This filter is a 15kHz lowpass (actually 4990 ohms and 0.022uF for rolloff above about 14.5kHz). The lowpass filter cuts low-frequency RF signals which the meter would otherwise show. The output of the lowpass filter exits the box.

The resulting amplified signal is parallel-connected both to an ancient Hewlett-Packard 3400A analog RMS voltmeter and an old stereo amplifier and speakers. It is crucial to be able to listen to noise before one starts to measure it, because various different problems may need to be solved first.

In section 7.21 of The Art of Electronics we find the equivalent noise bandwidth for RC-filtered noise. The bandwidth, B in hertz, for lower frequency f1 and higher frequency f2 (both single-pole RC filters) is:

   B = (Pi/2)((f2*f2)/(f1+f2))
Here we have f1 = 965 and f2 = 14500, so the equivalent noise bandwidth B is 21355Hz.

Terry Ritter, his current address, his top page and electronics home.