The principle of this experiment for generating sub-Poissonian light is the standard constant-current operation of a light emitting device [8, 9]. In order to suppress the number fluctuations of photons from the LED, we suppress the noise of the current injected into the LED. This can be realized using a current which flows through a high-impedance resister that is connected to a constant voltage source. The calibration of the shot noise level is done using a Poissonian current source [10]. In the case of a LED, it is difficult to use a balanced detector system for the calibration because a LED has wide radiation directivity in general. The circuit layout of the experiment is shown in Fig. 1. The LED is connected to a constant voltage source through a metal-film resister (Rs) or a photodiode (PD1) illuminated by white light. The LED can be driven by two kinds of current modes by switching from Rs to PD1. The LED generates sub-Poissonian light when it is driven by the constant current through Rs (SP mode), or it generates shot-noise light when it is driven by the current through PD1 whose electrons have Poissonian statistics (P-mode).
In this experiment, in order to observe the fluctuations of light from the LED, we have measured the fluctuations of the output photo-current from PD2. The input current to the LED is converted into light with efficiency of , and the photons enter PD2 with efficiency of and are converted into electrons with efficiency of . The ratio of the output current to the input current gives the total transfer efficiency which is the product of , and . The collection efficiency is high ( 0.9) because the LED is placed just in front of PD2, and the quantum efficiency of PD2 (Hamamatsu S5107 or S5106) is very high ( 0.96) in our experiment. Therefore, the emission efficiency makes the dominant contribution to the total transfer efficiency, and consequently to the amount of the noise reduction.
The output signal given as a voltage signal through is amplified by a high-input-impedance preamplifier (NF LI-75A, input impedance 100 M), and this is then fed to a spectrum analyzer (SA, Anritsu MS2601B). As described above, we peformed two sets of measurements; SP-mode and P-mode. In each set of measurements, we observe the SP-mode noise (or P-mode noise) and the amplifier noise power as a background noise. The amplifier noise power is subtracted from the SP-mode and the P-mode noise power, and then the normalized noise level is given by the ratio of sub-Poissonian noise power to the Poissonian noise power. In this experiment, because the differential efficiency is close to the total transfer efficiency, the normalized noise level can be regarded as the Fano factor [11]. The Fano factor should be equal to 1- under ideal conditions.
Figure 2 shows the total transfer efficiency of the LED used in our experiment (Hitachi HE8403). The upper trace is the efficiency at liquid nitrogen temperature and the lower trace is the efficiency at room temperature. At liquid nitrogen temperature, the LED shows quite a high efficiency. When the input current is larger than 30 A, the total transfer efficiency exceeds 0.5, and reaches 0.57 for an input current range larger than 1 mA. Even when the input current decreases to 1 A, the efficiency is still 0.32. This means that we can generate highly squeezed light at such a weak power level by using this LED. At room temperature, the efficiency becomes lower than that at liquid nitrogen temperature, but is still higher than that of the previously used LEDs [7]. When the input current is 10 mA, the efficiency is 0.41, and it decreases as the input current decreases. This decrease can be explained by considering the contribution of the generation-recombination current to the total forward current [13]. The reason for the superior performance in emission efficiency could be because (1) the chip surface of the hitachi LED is hemispherically shaped and (2) a metal layer between the p-n junction and the submount works as a mirror [12].