Clock Extraction using Two-contact Devices
Ch 5 Experimental Results
the ‘0’ level. This is the result of the nonlinear behaviour of the NLOA and detection of the output using a broadband detection system showed the characteristic behaviour described in Chapter 3 (eg. Figures 3.14 & 3.20). The electrical signal seen at the absorber contact contained a clock frequency component ~ 25-30 dB greater in power than that of the input signal, Figure 5.2b. The clock signal after the SAW filter is shown in Figure 5.2c and an identical signal was observed for a 2^®-l pattern.
a) (86.8MHz - 481.9MHz; lOdB/div) dominant peak @ 1 5 5 MHz
ck
k
• t , '
4'. f ■
b) (OHz - 200MHz; lOdB/div) c) (2ns/div)
Figure 5.2: a) Transmission spectrum fo r a 155.6 MHz 5 4 b) absorber contact spectrum fo r a 2^-1155 M bis data, c) extracted electrical clock signal from SAW output.
a) (20mV/div) b) (50mV/div)
Ch 5 Experimental Results
It was noted in Chapters 2 and 3 that the performance of the NLOA is dependent on the detuning of the input signal wavelength from the FP mode, eg. Figures 3.18 to 3.20. It is therefore expected that this clock extraction technique will be dependent on the input wavelength. The observed dependence is shown in Figure 5.4 which shows the measured device gain and the peak-peak clock signal level, Vp,p. Zero detuning corresponds to peak device gain and the onset of dispersive effects (cf. Figure 3.17). No clock signal was obtained for negative detunings and for small positive detunings (9f < - 3 GHz) the clock signal increased rapidly as the detuning increased. For detunings between ~ 3 GHz < 3f < 12 GHz Vp.p remained large, decreasing slowly as the device gain reduced. For detunings 9f > ~12 GHz the clock level reduced rapidly but remained at a low level for detunings of up to 9f ~ 19 GHz. This behaviour correlates with the observed nonlinearity in the output pulse shaping, see Figure 3.18. As the detuning increased the change in power at switch-on increased and the pulse becomes more rectangular - the extent of the nonlinearity increased. The clock component power level also followed this detuning behaviour with a high clock component level being observed for detunings 9f < ~10 GHz and lower levels for 9f > ~10 GHz.
1.8 ^ 1.6 9" 1.4 f
&
1.0" cn o 0.8 a 0.5 Gain clock Vp-p No 0.4 -5 0 5 10 15 20I
I
Detuning GHzFigure 5.4: Dependence o f device gain and extracted clock signal level with detuning o f the input wavelength around a FP mode.
From Figure 3.18 it is noted that the mark-space ratio of the output data also increases with increasing detuning and this change in ratio will result in an increased sensitivity penalty and is therefore not advantageous. It should be noted that clock component generation was achieved
Ch 5 Experimental Results
when the mark/space ratio was unity. In order to investigate the purity of the extracted clock, and establish whether the form of the output data resulted in a sensitivity penalty, BER measurements were compared using the extracted clock from the absorber contact and the transmitter clock signal. The results are shown in Figure 5,5.
I
a CQ 3 10 -4 10 ■5 10 6 10 -7 10 * no NLOA o Tx clock 10-'" 10 10"'^ + NLOA clock -48 -47 -46 -45 -44 -43Receiver Input power dBm
Figure 5 5 : BER variation with receiver input pow er fo r 155 M bis PRBS data.
The curve shown by (*) shows the BER variation in the back-to-back performance without the NLOA in the optical path and shows a 10’^ receiver sensitivity of -43.8 dBm. The small 0.7 dB penalty compared to the theoretical receiver sensitivity is primarily due to the on/off ratio of the modulated data (see footnote 5 in section 4.4). With the NLOA in the optical path the BER results when the BER receiver was synchronised with the transmitter clock is shown by (o) and those when synchronised using the clock from the absorber contact by (+). Both curves showed a 0.7 dB sensitivity penalty compared to the back-to-back measurements (*). This excess penalty arises from the lower on/off ratio at the NLOA output (penalty ~ 0.3 dB) and the change in mark/space ratio to 1.3 (penalty ~ 0.4 dB), see footnote 6. There was no observable difference in BER performance between the two sets of measurements and this shows that the
Ch 5 Experimental Results
extracted electrical clock from the absorber contact is indeed a good one and well synchronised to the input data. These results also show that the NLOA does not induce significant penalties to the system and is therefore suitable for consideration for use in optical networks.
When the pattern length was increased to 2^^-l then the BER results showed a small change in gradient but gave the same receiver sensitivity, see Figure 5.5. A 2^^-l pattern showed the beginnings of an error floor. This was probably due to the limiting Q of the SAW filter which resulted in periodic loss of clock signal during long lengths of ‘O’s.
The repetition capability of this method of clock recovery is limited by the repetition capabilities of the NLOA and the electrical filters available. We have seen in Chapters 3 and 4 that Gb/s operation is possible in both bulk and MQW devices. Previous measurements have shown that for the ridge type devices, like 18721 used here, the repetition capability is limited to ~ 500 Mb/s.
A i '' V V
f \ r-
V V
a) (OHz -1.5 GHz; 5dB/div) b) (Ins/div)
Figure 5.6: a) MQW absorber electrical spectrum fo r 1.4 Gbis NRZ data, b) extracted 622 MHz clock signal.
Indeed experiments using different SAW filters indicated that similar clock recovery was possible up to ~ 500 Mb/s. Using an MQW device, 21349, clock component generation at frequencies up to 5 Gb/s was obtained. Figure 5.6 a shows the absorber electrical contact spectrum for a 1.4 Gb/s 2^-1 NRZ pattern. As before, a strong clock component is generated. For the MQW devices the power of the clock signal at the contact was much reduced relative to the bulk devices. The low confinement factor for the MQW devices reduces the effect of the input signal on the carrier population relative to that for bulk devices. The absorber carrier density change required in MQW material is smaller than for bulk (due to higher gain coefficient) and so the shift in quasi-Fermi level (and hence contact voltage) is reduced. Figure 5.6b shows
Ch 5 Experimental Results
the electrical clock signal extracted from 622 Mb/s data using a 622 MHz SAW filter. Naturally the clock component is also generated in the optical modulation spectrum. Figure 5.7a shows the modulation spectrum of 2.5 Gb/s NRZ data injected into the MQW device and the output modulation spectrum is shown in Figure 5.7b. The input signal shows a suppressed clock component whereas the output spectrum has a strong clock component ~ 1 dB down in power from the low frequency data components. It is also evident from Figure 5.7 that the data signal has been amplified by ~ 4 dB. At 5 Gb/s similar operation was observed except that the clock component power was reduced by ~ 5 dB, probably due to the limiting repetition rate capability of this device. For higher input powers to the device ~ 7 Gb/s operation was observed.
avg Power -18,7 dBm avg Power -14.3 dBm ck
a) (OHz - 4GHz; 2dB/dlv) b) (OHz - 4GHz; 2dB/div)
Figure 5.7: a) Input modulation spectrum to NLOA fo r 2.5 Gb/s 2^-1 NRZ data, b) spectrum cfter NLOA.
'50 dB Monitor
Step-Recovery Diode
NLOA
Figure 5.8: Expt. arrangement fo r generation o f lOOps reverse bias pulses from extracted clock.
As described earlier in Chapter 3 it is sometimes desirable to use short electrical reverse bias pulses to quickly reset the switching device by reducing the turn-off time to <100 ps, eg. Figure 3.27. This clock recovery technique provides a simple way of implementing this function.
Ch 5 Experimental Results
Figure 5.8 shows the experimental arrangement for generating 100 ps electrical reverse bias pulses directly from the input optical NRZ data. The NLOA device used for these results was a bulk BH device 21327 which had a lasing wavelength of ~ 1.59 |im. The input wavelength was at ^ ~ 1.56 fim, the input power was 10 p.W and the spontaneous emission ripple was ~ 5 dB. This device showed nonlinear pulse shaping even at a bit-rate of 2.5 Gb/s. 500 Mb/s NRZ data was injected into the device. The absorber contact (Figure 5.9a) was filtered using a 500 Mb/s SAW resonator with a Q~8000 (loss - 1 2 dB). The electrical clock output from the SAW was split using a power divider, one signal being fed to a step-recovery diode and the other to the oscilloscope. A bias insertion tee at the input to the step-recovery diode enabled the DC bias to the diode to be adjusted for optimum pulse response. Various electrical amplifier and attenuators were used to obtain and maintain the desired power within the circuit without self-oscillation. Figure 5.9b shows the electrical clock signal (realtime scope) at the SAW output and Figure 5.9c shows the output from the step-recovery diode with 14 dB electrical attenuation. The electrical pulses from the step-recovery diode were very stable, had a good contrast and were of - 1(X) ps duration. The peak voltage of the pulses was — 3 V making these pulses very suitable for resetting bistable semiconductor devices as described in section 3.8 and in references 3.32-3.35.
A A
\j \j
0 UFN I UFS STORED
a) (centr&=500MHz,lMHz/div; 5dB/div) b) (Ins/div) c) (500ps/div)
Figure 5.9: a) detail o f absorber contact spectrum showing generated 500MHz clock component with adjacent data components, b) Extracted clock signal at 500 MHz and c) output from step-recovery diode
5.3.2 Discussion
This clock extraction technique is novel and has a number of possible application areas. The ability to extract a clock signal from optical NRZ data directly without sampling or detecting