All-Optical Regenerative Repeater

the last and unfinished work by A.H. Reeves

The real aim of telecommunication now is the virtual elimination altogether of the effects of at least terrestrial distances, not merely their alleviation. One consequence must and will be an almost universal picturephone service, for a multitude of purposes, that is lifelike enough to be generally accepted as an alternative to the much slower and more expensive human travel. Such necessary but outstanding picture quality will be impossible within the bandwidths of the present local areas, even by less redundant picture-transmission methods. There is no known reason why it should not become economically possible, though, by the use of the much greater bandwidths that optical methods can provide, not merely on the longer routes but also, by fibre optics, in the local areas as well.

It is my firm belief, therefore, that optical communication, suitably combined with other wideband methods such as millimetre waveguides, will eventually become the main backbone of the world's networks. To check such beliefs, due research expenditure now is thus justified on not merely the first generation of optical devices such as repeaters, but also on more advanced designs that should give a truer comparison between the various wideband methods at a later date. Let us now see why such more advanced repeater designs are necessary.

Optical circuitry is in its infancy. It is for this reason that all optical repeaters note being developed for early use, as far as I know, are (in radio terms) still in the era of the "crystal set", with no RF of IF gain. During the next ten or twenty years, though, this situation will change completely; practical optical equivalents will be found for almost all the electronic tools that are now so familiar, including gates, level clippers, waveshape regenerators, retimers, and many others. Because of the much reduced wavelengths, some of them, digital circulating delay-line stores in particular, will be smaller and cheaper than their baseband counterparts. quite a few will first rind application not in telecommunications but in the high-speed computer field, this second art thus helping and speeding the first. A start has already been made, especially in high-speed optical gates. The present Study Contract is a further attempt to open up this new and exciting field.

In the usual designs, the received optical signals are applied to a photo-diode, the resulting baseband output waveform is amplified, re-shaped and re-timed by fairly conventional PCM-type circuitry, amplified further and caused to modulate a GaAs laser which is applied to the next repeater section of the fibre or other optical waveguide. The concept is straightforward and simple; and all components except the laser can be of proved reliability. As the optical detector has a very broad band, there are no optical tuning problems.

The main economic drawback of this simple repeater, though, is that with present-type components the bit rate is likely to be limited now to about 0.5 Gb/s, with perhaps 1 Gb/s as the ultimate figure. The first limiting factor here is the maximum speed of the re-shaping and re-timing circuit. 0.5 Gb/s can hardly cater for even two PCM-ed picturephone channels of the future needed standards (about 300 Mb/s per channel). Moreover, even if the photodiode is correctly matched at 1 Gb/s to its output line, extrapolation from present devices gives a noise figure of about 16 dB compared with basic photon-arrival noise (for an input analogue signal-to-noise ratio to give a digital error rate not exceeding 1 in 1010 per repeater). 16 dB is not ridiculous; but it is by no means good.

On the longer routes the repeaters, probably not more than one to three kilometres apart, will be a major factor in the total cost. One way of minimising this expense, is, of course, to cater for as many channels as possible in each repeater, giving a bit rate in each fibre, carrying one TDM group, that is really maximised. If we can aim at, say, 10 Gb/s per repeater, while catering initially for rates down to about 1 Gb/s, we are tackling the system economics from a much better starting point. An answer, of course, is to amplify and process the signals not at baseband but at a carrier frequency where the highest desirable signal components are only small fractions of this frequency. A superhet receiver, though feasible, would not give any obvious advantage, as an IF amplifier design of, say, 100 Gb/s midband frequency and 20 GHz wide has at present no convenient solid-state solution (and a TWT, with its high voltage and vacuum drawbacks, would in any case hardly be acceptable to the British Post Office in this position).

By this "reduced percentage bandwidth" approach it would be much better, therefore, to go all the way and do all the amplifying and processing at the optical frequency itself. For the amplification a direct laser amplifier would be possible. But as we may need repeater gains, eventually, of at least 60 dB it would be quite a problem in practice to ensure that this amplifier does not oscillate. Any positive feedback reflections would have to be at least 80 dB down - almost impossible to ensure even by the best "blooming" (optical impedance matching) methods. Even if a "Faraday window" were added to help the unidirectional action by, say, a further 20 dB, an undesirable and probably rather bulky complication, the remaining 60-dB-down limit on the reflections would present in a practical design an almost impossible problem.

In my opinion there is, though, a practical answer. It is the use, at optical frequencies, of what used to be called the "Fluelling" modification of a super-regenerative receiver, by which the quenching waveform is periodically self induced, rather than being applied externally. As will be shown later, we can use this circuit not as a complete receiver (with built-in RF gain), which was its original purpose, but as & stable optical amplifier. By using correctly a further built-in feature of the circuit we can obtain our needed top and bottom level clippers. And in at least two different but simple ways we can add the necessary pulse re-timing.

Here in the original study, C.C. Eaglesfield attempted to elaborate on these ideas. This may be published here at a future date when time permits. The following section by A.H. Reeves was added to the study as an appendix.

It is well known that a self-quenching super-regenerative device contains an oscillator, usually of nearly sinusoidal waveform, having an amplitude envelope of pulsating waveform. It is also well known that to achieve this result the oscillator during each pulse period must have a suitably varying damping factor. For self-quenching this damping factor must be produced internally, being a combined function of oscillator amplitude and time. Let the instantaneous voltage of the oscillator be represented by Y.aeat.sin wt. The damping factor is therefore a.

The exact times of arrival of the voltage pulses will depend on the amount of noise collected. If we assume this noise to have a random time and amplitude distribution there will thus be a random time fitter on the leading edge of the voltage pulse. As will be explained later, in the self-quenching mode the timing of the damping waveform is dependent mainly on the timing of the leading edge of the voltage waveform shown in Fig.8.1(b). There will thus be a time filter on the complete voltage waveform as shown for example in the dotted curve of Fig-8.1(b).

We shall now explain how in the self-quenching mode the required damping waveform of the type shown in Fig.8.l(a) is derived from the voltage waveform of the type shown in Fig.8.1(b). In the usual electronic versions, the peak voltage depends on the saturation level of the active device giving gain that is included in the positive feedback loop of the oscillator. At the saturation level the net gain round the feedback loop averaged over one period is zero. To understand the rest of the damping waveform an important basic point must be noted. It will be seen that the shape of the damping waveform of Fig.8.1(a) is not symmetrical with respect to the point at which the voltage waveform of Fig.8.1(b) reaches its peak. If it were symmetrical there could be no pulsating voltage envelope, as in that case when the voltage waveform reached the saturation level, corresponding to a net damping of zero averaged over one complete period at the oscillator frequency, the oscillation amplitude would remain at that level indefinitely. To produce the needed voltage envelope pulsations the peak of the positive damping waveform must be suitably delayed with respect to the positive peak of the voltage waveform. In electronic devices the positive damping peak is usually obtained by rectifying the oscillatory wave, allowing the rectified output to charge a condenser, and applying this charge as a back bias on the oscillator device itself in such a way that this active device then has zero gain, the resulting damping then being that of the resonant circuit that determines the oscillator frequency. To meet the above delay requirement of the positive peak of the damping a suitable delay is added to the back-biasing circuit from the condenser on which the charge is built up.

If the active device included in the oscillator feedback loop operates as a class B amplifier the net gain averaged over one oscillation period will be less when the oscillation starts to build up than at moments later on, until saturation begins to set in. This can cause the envelope to be bi-stable in level (though without a suitable delay on the positive damping peak the pulsating envelope cannot be obtained). When the suitable delay is added, however, to the class B operation, narrower pulse widths with sharper rise and fall times can be obtained. This can be a useful improvement. In optical versions it can be particularly important.

An approximation to the oscillatory response can be obtained by simple conventional circuit design using four CRL branches in combination with normal diode non-linear devices - though rather more sophisticated design is needed to give adequate practical tolerances. Unless the active device giving gain in the oscillator positive feedback loop operates in class B, the positive feedback loop must have two separate branches attached from two different paints on the network.

We shall now examine the response of this self-quenching super regenerative (SQSR) device to a single signal pulse with the envelope shape shown as A in Fig.8.1(c).. The saturation voltage will be nearly the same as before. The waveform of the voltage trailing edge will be unaffected by the signal pulse but it will now fall to zero earlier than before as shown. The result will be a slightly narrower pulse than in the absence of the signal.

If now a single signal pulse of the same strength as before were applied but slightly retarded in timing as shown at B of Fig.8.1(c), the response of the device would be as shown in Fig.8.1(d), the peak of the voltage pulse being still advanced beyond that due to maximum noise alone but less so than after A.

If during the period of negative damping that damping were precisely constant the time spacings between the output pulses of the repeating device would be exactly the same as that of the input. As such constancy is not practicable, however, there will be a slight error in reproducing the timing, which can become cumulative on any one pulse as it progresses through a subsequent chain of repeaters. This cumulative error can be corrected by superimposing a slight oscillatory component as shown in the hatched line of Fig.8.1(a). It should have a maximum slope in the negative direction at the peaks of both the A and the B pulses shown in Fig.8.1(d). The effect is to stabilise the pulse timings at the output of the repeater in the positions required for reproducing exactly the time spacings at the input, irrespective of the number of repeaters.

In a digital modulation system such as PCM, pulses at the timings shown at A and B can represent the one's and nought's respectively of the code used. It will be seen by the relative timings of the curves shown in Fig.8.1 that the system works correctly whatever the sequence between the one's and nought's that is used.

Optical SQSR repeaters

To reproduce sharp pulses the first requirement is a high gain per wavelength as an optical amplifier, in order for the leading edge of the output pulse to build up fast enough. The GaAs variety of the semiconductor laser group best fulfils the requirement at the present time. The second requirement is a fast enough fall time from the repeated waveform. With the usual junction type having a d.c. drive it may be difficult to get the peak of the + a large enough to damp the voltage peak fast enough. But there is an alternative, to reduce the volts after the positive peak at the required speed by reflecting back an out-of-phase optical wave-train derived from the rising voltage train itself.

To achieve SQSR operation all we need to do is to apply the same principles as in the normal electronic examples already explained. In junction-type lasers, though, at the present time, if we want an envelope p.r.f. exceeding about 1 GHz it is difficult and usually uneconomic to change the damping of the laser oscillator circuits fast enough to prevent the envelope rise and fall times being limited by the bandwidths of the current waveforms that drive them. Instead of copying exactly the electronic devices, in which the oscillator is back-biased by a suitable self-produced base-band wave, it is more convenient to obtain the required damping waveform by arranging to produce, from the envelope waveform of the laser output, a suitable and suitably delayed change in the extent of the "inverted energy population", according to now well-known quantum-mechanical laws. To do so two separate and similar laser devices may be suitably coupled together - alternatively, the two lasers may be made from the same slice of gallium arsenide.

The (steady) current drive to laser (1) is enough to give adequately fast envelope rise times during its lasing build-up period. The drive to laser (2), however, is smaller - its junction length optically attenuates, due to the inverted population from its d.c. drive alone.

To explain its notion, assume that an optical signal or noise component causes laser (1) to build up an oscillation.

Until laser (2) receives input from laser (1) it attenuates sufficiently to prevent the energy in its reflected wave back to laser (1) from becoming enough to influence appreciably the laser (1) waveform. But again by well known laws of quantum optics, any group of photons impinging on laser (2) from laser (1) that has an energy level per photon of (E2 - El) causes, in any group of electrons in laser (2) Junction where (as in the present case) quantum laws allow an energy rise in these electrons by the amount (E2 - E1), a rise by that amount in those electrons that "capture" those photons that are at the lower level of these two energy states - while these laser (2) electrons that capture the photons but are initially at the higher energy level emit a photon and lose energy. Initially, as laser (2) is attenuating, there are more electrons in the lower states than in the higher. Where the numbers are exactly equal the junction material is statistically transparent.

In the present case, though, there is a statistical energy-level rise, - towards transparency in laser (2) junction. Let the d.c. drive on laser (2) be such that in time enough photons from laser (1) have been captured by the laser (2) lower-state electrons to render laser (2) junction nearly transparent.

Eventually the inverted energy population in laser (1) will have been used up to an extent well below the threshold for lasing, The laser (2) junction has also used up most of its inverted population, due to its input from laser (1), rendering it relatively opaque.

The laser (1) oscillation and therefore also its effect in making the laser (2) Junction more opaque, will stop - until such time as is necessary for the d.c. drive on laser (1) to restore enough population inversion for its oscillation to restart.

We thus get a recurring envelope pulse giving an overall pulsating envelope waveform of a duty cycle that is controllable, in particular by the amount of d.c. drive to laser (1).

If now a suitable signal is applied, of an optical carrier frequency within the receivable bandwidth of the device, with a suitable envelope shape of a p.r.f. suitably higher than that of the optical SASH device when the signal pulses are absent, the device will act as a digital repeater for the signals in the same way as already explained for the normal electronic versions.

With digital PTM, to avoid cumulative retiming errors as an envelope pulse travels through a chain of repeaters, a suitable, small oscillatory waveform must be added to the damping wave, as already explained and shown in Fig.8.1(a). It may be obtained by tapping off some of the repeatered optical output into a photo-diode, for example, and suitably processing the resulting base-band waveform by conventional means.