The circuit layout of the experiment is shown in Fig.1.
Figure 1: Experimental Arrangement. By switching from the resistor Rs to
the photodiode, sub-Poissonian and Poissonian noise power can be
measured.
An LED (Hamamatsu L2656) is connected through a metal-film resistor (Rs) to a filtered power supply. This model of LED is used in the most of the previous experiments[4, 5, 6, 7, 17] except the first experiment by Tapster et al.[16]. As the power supply, we use a commercial voltage regulator (MTR 18-1), or at weak current level, we use a conventional dry battery because we can not eliminate completely the electric noise that might come from the ground contact. The mean current is determined by measuring the voltage drop across the resistor Rs. In order to calibrate the shot noise level, we generated a Poissonian driving current by using a photodiode illuminated by white light as a stochastic regulator. By switching from the resistor Rs to the photodiode the experimental arrangement can be operated in sub-Poissonian or in Poissonian mode without modifications of the detection geometry [17]. In order to generate white stochastic noise, viz. the noise spectrum of the current being flat, the response time of the photodiode must be uniform over the frequency range of interest.
The LED is placed just in front of a PIN photodiode (Hamamatsu S5107,
quantum efficiency 
0.9) in a Teflon housing.
This face-to-face arrangement allows a high light collection efficiency
( 
0.9).
It was operated at room temperature and at liquid nitrogen temperature.
The output current of the photodiode is converted to a voltage signal
through a load resistor 
. The response time of the detection system
is mostly determined by the RC low-pass circuit formed by the
capacitance of the photodiode and the load resistor. The voltage signal
is amplified by a high-input-impedance preamplifier (NF LI75A, input
impedance 
100M 
), and this is then fed to a spectrum
analyzer (SA, Anritsu MS2601B).
The SA operates for frequencies above 9kHz.