With transmission systems demanding data rates in excess of 150 GBit/s (see table 1.2)
pulse requirements are for very short duration with high frequency components. The high gain high bandwidth of TWSLAs compared to FP devices is particularly suited to this application. In this section the amplification of pulses with a width of a picosecond or greater are examined. The special treatment required for subpicosecond pulses follows in a later section. The use of TWSLAs for pulse processing has led to a number of applications for their use, e.g. optical sampling [Jinno et al 94], non return to zero (NRZ) to return to zero (RZ) format conversion [Noel et al 95] , logical operations (AND gate) [Nesset et al 95] and address recognition [D’Ottavi et al 95], These examples highlight the increasing use of an optical amplifier to perform some of the functions on optical signals which were previously the domain of electronic methods.
3.5.1 Pulse response TWSLA (Experimental observations)
3.5.1.1. Regular pulse input at rates > recombination timeInitial investigations of TWSLAs centred around a steady state time response. The first measurements on the dynamic response of a TWSLA were made by Marshall [Marshall et al 87]. The pulses used there had a FWHM of « 50 ps which were amplified at low rates (100 MHz) to a peak power of 100 mW. Antireflection coatings were applied to this device which gave facet reflectivities of « 0.08%. Even at this low level of reflectivity TWSLAs still exhibit residual cavity modes, in this case ripple depth was 3 dB of the modal gain which peaked at 25 dB. The input consisted of regularly spaced pulses (10 ns apart). Increased pulse repetition rates (2.5 and 5 ns spacing) gave identical gain curves as the 10 ns case. In all instances the pre-pulse carrier density
settles to the zero input value. Further investigations in an experiment by Wiesenfeld [Wiesenfeld et al 88] used pulses considerably shorter (7.9 ps ) than before. The pulse separation was above the relaxation time (i.e. a pulse repetition rate of 200 kHz giving a time of 5 ^is between pulses). In this case the pulses were characterised by their energies rather than powers. This is more appropriate for widely separated pulses
whose effect on the carrier density is independent from adjacent pulses. Pulse
saturation power would be misleading with different average pulse powers giving different values of gain saturation. The amplifier device gain is measured against a value of energy saturation which is the value of a single pulse when the gain begins to reduce. Distortionless pulse output was observed when the condition Ein < Esat (where Esat is the saturation energy given by Psa/x2) was satisfied and no gain saturation occurred. With Ein approaching Esat gain saturation was apparent. The pulse input width for increasing energy was kept constant for this experiment therefore increasing the input energy is equivalent to increasing the pulse peak power. The gain v output energy curve was described as linear over a region where Ein « Esat with a response given by [Wiesenfeld et al 88]:
G = Go - 4.3 4Eout/Esat. (3.22)
Further experiments on pulse amplification were carried out by Eisenstein [Eisenstein 88] at repetition rates of 4 GHz and 1 GHz where gain compression occurred for the same pulse energy in each case. The recovery time was 250 ps for this amplifier giving independent pulse amplification for both frequencies. Using data rates around the reciprocal of the relaxation time ensures no intersymbol interference occurs between adjacent pulses as each pulse experiences a gain which has fully recovered from the previous pulse, giving a maximum data rate of 4 GBit/s in this case. Figure 3.8 shows
an amplified 4 GHz distortion free pulse which would result in no intersymbol interference [Eisenstein et al 88]. IN PUT PULSE 0 0' 10 AM PLIFIED PULSE 12 ps 0 0, 20 4 0 TIM E ( p j ) €0
Figure 3.8 Distortion free pulse at low data rate [Eisenstein 88].
The point to note in the above experiments is that operation at data rates higher than 1 /t2
produces identical gain for each pulse even under gain saturation. 3.5.1.2 Modulated pulse input in saturation region
The previous experiments used a periodic pulse input. An interesting feature of dynamic input to TWSLAs is when modulated data is input to the amplifier. It is observed that the power gain can increase momentarily when a signal contains a train of ones interspaced by zeros. The experiment of Ligne [Ligne et al 90] satisfied this condition with pulse rates at 1 Gbit/s and 4.8 G bit/s with r2 = 2 ns. A pulse train consisting of 011111010 was input to the amplifier. Common to both data rates the amplifier output was non linear with significant overshoot occurring on the pulse following a series of zeros (bit 2 and 8). Stable conditions of the amplifier were achieved after a series of Is (bits 2 to 6), ( figure 3.9).
Figure 3.9 High data rate pulses to saturated TWSLA [Ligne 90].
With the lower data rate (1 Gbit/s figure 3.9a) the long period of the first bit allows for stability conditions to be reached over a one data pulse and a longer period for carrier relaxation when a zero is received. Figure 3.9b for a data rate of 4 GBit/s shows the second overshoot is lower than the first, the reason being that the short pulse period requires a larger number of pulses for stability to be reached. If the amplifier was operated in the unsaturated region all bits would receive equal amplification for all data rates. The TWSLA when operated in the saturated region gives an uneven response for different bits of the signal due to the variation in carrier density associated with the saturation region. Similar experiments by Inoue and Yoshino [Inoue and Yoshino 96] have showed pulse distortion is evident for modulated bit rates (7.5 Gbit/s) under gain
saturation of the device, see figure 3.10 which shows pulse distortion in the closed eye diagram.
Figure 3.10 7.5 Gbit/s pulse amplification in saturated device [Inoue and Yoshino 96], In the modulated data rate case it is clear that consideration must be given to the effects of saturation and the timing between data pulses, although the average gain of a train of pulses may yield a certain value a different value occurs for different bits.