Smooth white noise source in the range of 1Hz~100kHz

White noise is very useful for testing many different types of circuits. When used in conjunction with an FFT analyzer, a smooth noise source can help to quickly and easily generate a circuit gain layout plot. If the noise in a circuit is smooth and the amount of noise is known, the gain of the output circuit can be easily determined, even visually. This method was used as early as 1978 on the HP3582A low-frequency spectrum analyzer (Reference 1).

A "modern" way to generate white noise is to use a digital shift/feedback register arrangement in a CPLD or FPGA. Some people have even created a parallel array of microcontrollers for generating Gaussian white noise.

The following implementations are all simulations, and if necessary, some components with through holes can be used to make the installation of prototypes easier.

It is well known that Zener diodes are a good source of broadband noise. As always, the trick is to find a diode that is relatively smooth over the desired frequency range. Jim Williams used a common 6.8V Zener diode in his 5MHz broadband noise generator. It is also common to reverse bias the base-emitter junction of an NPN transistor and use it as a noise diode. The goal of this Design Idea is to generate a large amount of relatively smooth noise over the frequency range of 1Hz to 100kHz for FFT testing.

Using a 6.8V Zener diode does produce broadband noise, but there is a lot of 1/f noise at low frequencies and it is not smoothed out to DC. So in this design, reliable 12V Zener diodes were used as noise sources. These diodes are smooth over frequency, produce a lot of inherent noise, and work well over the discharge life of a 9V battery (Reference 2).

When the diode used is biased to 18V, it generates about 20mVRMS of intrinsic noise through a 1MΩ resistor. This scaling is fortuitous, as the peak-to-peak value is about 5 times that value—100mV.

To provide a continuous check of the DC offset error, a vintage LF412 dual JFET input op amp is used to amplify the diode noise in two x10 steps.

Figure 1 shows the circuit that was built. The biasing of the Zener diode occurs from two 9V batteries (18V) in series through a 1MΩ resistor. The LF412 operates from a single supply between the two 9V batteries. With low input current and low offset voltage, bulky output coupling capacitors can be eliminated because the DC value of the output is within millivolts of ground. Although the LF412 is not a low-noise op amp, its noise level is still far below the intrinsic noise level of the diode, so this is not a concern.

U1A and U1B amplify the Zener noise to approximately 1V and 10V p-p in two x10 steps. If this noise is too much for the circuit under test, 1kΩ R8 and R9 can be used to create a voltage divider to reduce the noise level to the desired value.

Figure 2 shows the noise generated by the circuit, which is smooth from 1Hz to 100kHz. The higher 100kHz frequency attenuation is less than 0.5dB in the x100 output. If necessary, frequency-dependent gain can be added to U1B. However, for this experiment, it is not necessary. For comparison, Figure 2 also shows the noise layout of the LM317 regulator running at minimum capacitance. Although this device is generally considered a "very noisy regulator", it is not worth mentioning compared to a 12V Zener diode and some gain.

By using a stable low-power FET amplifier and ceramic coupling capacitors, the 1/f noise caused by the temperature gradient of the discrete air flow is suppressed to a minimum. However, for maximum stability, this design example should be performed in an enclosed environment and away from circulating air flows.

The circuit consumes only 4 mA and the life of a common 9V alkaline battery is expected to be greater than 100 hours when the battery is used up to 7V. The noise variation of the designed circuit is about 15% of the battery life, and this can be improved by using a more complex and stable bias arrangement if necessary.