**5. Discussion**

The transfer functions shown in **Figures 9** and **12** comprised for each demultiplexer channel:


The loss of the connector interface was minimized by applying the index-matching gel. This loss downscaled the transfer function of the demultiplexer by approx. 0.5 dB. To obtain the performance of the optomechanical setup itself, the value of 0.5 dB should be added to the measured transmittance values. The propagation losses included the Fresnel loss at the end face of the ingoing and the outgoing fiber, the losses introduced by the optical components (including reflections on the anti-reflection coatings), and the coupling losses due to the setup misalignments, optical aberrations, and clear aperture of components. For the perfectly aligned components and for given distances between them (obtained, e.g., from the CAD model), the minimum loss of the demultiplexer could be estimated by means of an optical ray tracing software. However, that work was beyond the scope of this work.

The shape of the spectral response of each demultiplexer channel was predominantly determined by an interference filter that was used. Those filters provided flattop response, steep transition slopes, and high isolation between the channels due to an optical density greater than 4 (transmission of <0.01%) in the rejection bands within 400–700 nm region. The deviations of the channels from the nominal central wavelengths and bandwidths of interference filters comply with the center wavelength and passband bandwidth tolerances of ±2 and ±5 nm for 10 and 50 nm filters, respectively. An exception is the green channel where the spectral response curve was truncated by 505 nm cutoff dichroic mirror with the transmission band starting at 520 nm.

The four-channel demultiplexer introduced an additional channel in the short wavelength region. That allowed simultaneous operation at the violet and blue wavelengths, which are both very attractive for POF communication due to the availability of commercial laser diodes. Two different demultiplexer configurations offered significantly different performance.

In the serial configuration, the longer wavelength channels corresponded to the higher output ports. Because of the longer optical path than the shorter wavelength channels, the longer wavelength channels:


The influence of those effects can be observed in the transfer function from **Figure 12**, where the green and red channels experienced significantly higher IL than the violet and blue ones. If the effect of alignment inaccuracy, which is a parameter related to the particular setup adjustment, would be disregarded, all other effects that are inherent to the serial configuration would lead to the same principal behavior of the transfer function.

The Appelt et al. [26] demultiplexer outperformed the four-channel solution from [28] in terms of IL and especially crosstalk. An exceptional performance of that demultiplexer with IL between 3.19 and 5.66 dB (overall minimum IL of 16.87 dB) may be explained by a very precise alignment of the components. However, all other measurements (performed 2 years thereafter) with two-, three-, and four-channel setups, which had to be each time newly aligned, showed somewhat higher IL but also very consistent behavior to one another. Therefore, it cannot be excluded that some other factors such as accumulated dust on the optical surfaces or coating damages due to improper handling could have introduced additional attenuation compared to [26] measurement, which was performed with brand new components. In spite of that, all subsequent measurement results, including the IL of 3.85–6.15 dB for the reassembled two-stage demultiplexer, can be considered as excellent achievements.

The significance of these and of the other previously realized interference filter-based SI-POF demultiplexers is that they enable realization of POF WDM systems and investigation on their data-carrying capacity. For these reasons it is important to further optimize the realized demultiplexer setup and extend the channel count.
