The noise levels directly observed with the SA are shown in Fig. 3. These values are those at the liquid nitrogen temperature when the output PD current was 1 mA (the estimated light intensity was 1.8 mW and the input current was 1.75 mA). The upper two traces show the noise levels for the P-mode and the SP-mode. At 400 KHz, the P-mode noise level is approximately 14 dB above the amplifier noise level as expected from the calculation (). There is a 3.2 dB noise reduction from the P-mode to the SP-mode noise power. The sharp peaks at 810 KHz and 954 KHz are due to the radio waves from the radio stations.
The normalized noise level obtained from the data in Fig. 3 is shown in Fig. 4. We plot similar experimental data obtained using another type of LED for comparison (Hamamatsu L2656, this LED has been used in most of the previous experiments [7]). The noise reduction by HE8403 is almost uniform within our measurement bandwidth of 1 MHz. The value of the normalized noise level is 0.40.47, and conforms with the measured total transfer efficiency of 0.57. This, to our knowledge, is the largest squeezing in photon number fluctuations ever observed using a LED.
An experimental result for a broader frequency range is shown in Fig. 5. To extend the measurable frequency range, a different preamplifier is used (NF 5305). The LED used in this experiment is a different sample but the same type that in Fig. 3. The result was obtained at room temperature and the output PD current was 500 A (the estimated light intensity was 1.8 mW and the input current was 2.6 mA). From this data, it is shown that this type of LED suppresses the photon number fluctuations uniformly over a frequency range up to 10 MHz. According to our recent experiment, the amount of the noise reduction decreases in a higher frequency region due to the collective Coulomb blockade effect and the contribution of the radiative lifetime of the LED [14].