Analog processing

For the analog processing approach with MATLAB the four QC output voltages are added, as depicted in Fig. 8. This sum is then filtered by a moving average filter to smoothen the signal. Finally, a decision threshold is used to obtain a logical “0” or “1” at defined sampling points.

Analog processing of the 4 quencher circuits (QC) output data for 50 Mbit/s RZ at an optical power of 7.5 nW. The signals for a sequence of 50 bits were exported from Matlab. The signals Q1, Q2, Q3 and Q4 were measured simultaneously with the oscilloscope and imported into Matlab; their sum “Sum” was calculated by Matlab. The “Filtered” signal was scaled by a factor of 2 to make the curve progression better visible. The dashed line represents the threshold used by Matlab. The final output data (“Out”) are shifted by half a bit period to the right compared to the input data “Data” because the decision is made in the middle of each bit (denoted by the circles). The input “Data” sequence is shifted by half a bit (i.e. by 10 ns) to the right and repeated as “Data Shifted” to make a better comparison of output data “Out” and input “Data” possible.

The window lengths of the moving average filter were 61 (61 samples at 5 GS/s correspond to a length of 12 ns; the duration of a bit at 50 Mbit/s is 20 ns) for the return to zero 50 Mbit/s measurements, 91 (corresponding to 18 ns) for the non-return to zero 50 Mbit/s measurements, and 51 (corresponding to 10 ns; the duration of a bit at 100 Mbit/s is 10 ns) for the return to zero 100 Mbit/s measurements. In a hardware realization the moving average filter can be replaced by a simple low pass filter and the decider can be implemented by using a comparator. It should be possible to further optimize the analog processing approach by utilizing more advanced filter topologies. BER was obtained in Matlab by comparing “Out” with “Data shifted”, counting the number of different bits as errors and dividing by the total number of bits.

To show the symbol-dependent hysteresis the eye-diagram shown in Fig. 9 was constructed from the filtered sum latches signal obtained by Matlab. The moving average filter using a length of 12 ns was implemented in Matlab. Four (not counting the zero line as a fifth) bright horizontal lines are visible in Fig. 9, which are caused by the addition of the four quenchers’ digital output signals. Within an eye of about 0.15 AU (arbitrary unit) for a duration of about 5 ns (between about 18 ns and 23 ns), the density of traces is lowest between the third and fourth bright line. The decision threshold of a comparator has to be set to the middle between the third and fourth bright line. This eye is large enough for applying a clocked comparator successfully. The quality of the eye is limited by the BER of 4 × 10⁻⁴ and the analog processing method used. When considering the trace “Filtered (scaled)” in Fig. 8, the limited quality of the eye is not surprising. There is, however, room for further work to find a better method of analog processing of the quenchers’ output data.

Eye-diagram for analog processing of the 4 quencher circuits (QC) output data for 50 Mbit/s RZ at an optical power of 7.5 nW. The filtered signal of the sum of the 4 latch outputs from Matlab was used to construct the eye-diagram. The clock from the bit pattern generator, which generated the Data signal was used as trigger to overlay 90.000 bits.

A better sensitivity was obtained with an optimized decision threshold for different optical input powers. Figure 10 shows the optimum threshold values used for processing of the quenchers’ output data. Such an adaptive decision threshold can be realized with a dedicated analog circuit or a digital counter plus a digital to analog converter setting its output voltage in an appropriate manner in dependence on the input optical power or count rate of the QCs, respectively.

Optimized decision threshold over average optical input power for analog post-processing.