High optical power drives 40-Gbit/s receiver design


by Roy Howard and Don Becker

To reach high-speed transmission, the design of optical receivers requires consideration of the method of amplification and the data format. High-quality eye diagrams are possible in both RZ and NRZ formats.

Moving to high-speed 40-Gbit/s systems presents challenges in the design of some optical components, including transmitters, fiber transport, and receivers. Recent developments in high-speed-receiver design include the use of ultrawideband preamplifiers and the use of novel materials on monolithic chips.

Receivers, in particular, will require several critical design changes for return-to-zero (RZ) and non-return-to-zero (NRZ) data formats. A critical first link in receiver design is the optical-to-electrical (O/E) converter. Indium gallium arsenide (InGaAs) PIN diodes convert packets of photons to electrons, allowing preamplification of either the optical or electrical signal. Ultrawideband electrical preamplifiers operating in the range of 30 to 40 GHz with 8 to 16 dB of gain, and 4 to 6 dB of noise are just now coming onto the market. These amplifiers are pushing the state of the art in semiconductor design.

A number of technology trends are under way. Silicon germanium, GaAs, and indium phosphide are used in the pursuit of low noise and high gain in monolithic microwave integrated-circuit chips. The design of broadband transimpedance preamplifiers that maintain flat gain and linear phase from 30 to 40 GHz is complicated by the nature of the millimeter-wave electrical signals that travel 1000 times slower than the speed of light. Even with these difficulties, such devices will proliferate.

Optical amplifiers, by comparison, are a mature technology; moving to 40 Gbit/s presents no new technical challenges. Erbium-doped fiber amplifier (EDFA) technology, for example, has seen steady improvement and cost reductions over the last decade. The introduction of Raman amplifiers—a complementary technology—will enable future lightwave systems to use a combination of EDFA and Raman optical amplification. Optical amplifiers have evolved for bandwidths from zero to well over 200 GHz, gains greater than 20 dB, and only 3 to 5 dB of noise.

Sources of noise
The noise limitations for optical and electrical amplification are significantly different. Pure electrical amplification is limited by the Johnson noise (electrical resistance-current noise) integrated over the bandwidth of the amplifier. For a 40-GHz bandwidth, sensitivity is limited to -12 dBm. Optical amplification is limited by the shot noise in the signal itself and any excess autospontaneous emission (ASE) from the amplifier. A large portion of ASE can be removed by optical bandpass filtering. Sensitivities for optically amplified systems should be in the range of -28 dBm, a 16-dB improvement over electrical amplification alone. In practice, the two noise sources add as the square root of the sum of the squares; either electrical or shot noise will dominate, depending on the optical power level into the PIN detector. When running at 40 Gbit/s, PIN detectors will become shot-noise-limited at an input power level somewhere between +3 and +6 dBm, depending on whether the data format is RZ or NRZ.

Moving from 10 to 40 Gbit/s will place an increasing importance on optical signal preamplification as a result of gain, bandwidth, and noise limitations of electrical preamplifiers. In the absence of electrical postamplifiers, InGaAs PIN diodes will need to handle up to +12-dBm average optical power at a 40-GHz bandwidth to drive clock- and data-recovery circuits in RZ format 40-Gbit/s receivers. For a combination of optical and electrical amplification, +6-dBm optical power is necessary to achieve shot-noise-limited performance. In either case, the O/E converter must handle high peak optical power at high bandwidth with a minimum of jitter. Ideally, the receiver will be polarization insensitive because polarization-dependant loss (PDL) would manifest itself as amplitude noise for a randomly varying input polarization signal.

An NRZ 40-Gbit/s pulse occurs over the entire "1" bit window, which has a time duration of 25 ps. The frequency response of this NRZ pulse is the classic sin(x)/x response of a boxcar pulse (see Fig. 1). Nodes occur in this response at integral multiples of 40 GHz, representing the frequency spectrum of the highest-frequency NRZ pseudorandom bit sequence (PRBS) stream obtainable—that of alternating "1" and "0" bits. A typical NRZ PRBS bit stream may have many strings of consecutive "1" bits of varying duration. The frequency spectrum of the typical NRZ PRBS bit stream will be reshaped toward lower frequencies. The -6 dB point for the alternating "1" and "0" bit stream occurs at 32 GHz. Thus, a bandwidth requirement of 32 GHz for a typical 40-Gbit/s NRZ system should give good performance.

An RZ 40-Gbit/s pulse can take on varying durations and shapes within the 25-ps bit window. A typical RZ pulse can have a duration of 80% of the total bit window. Assuming the RZ pulse is rectangular in shape, the frequency response of this pulse will be stretched out by a factor of 1.25 in frequency. The first node of the response will occur at 50 GHz, and the -6-dB point will occur at 40 GHz. Because the RZ pulses always return to zero, the exact nature of any RZ PRBS bit stream will not be pattern dependent, or word-length dependent. Typical 40-Gbit/s RZ systems will therefore require 40 GHz of bandwidth for good performance.

Requirements for RZ and NRZ formats
Key specifications for any optical receiver to achieve the maximum system-link budget are 3-dB electrical bandwidth, jitter, maximum optical input, and PDL. These specifications applied at 40 Gbit/s are quite different, depending on the data format selected.

For 40-Gbit/s systems, RZ data format requires more bandwidth, typically 40 vs. 32 GHz required for NRZ data. The higher bandwidth requirement for RZ formats is counterbalanced to some extent by the more forgiving jitter requirement for RZ formats. Return-to-zero and NRZ data require typically small jitter values, with 3.5 ps of jitter required for RZ formats and 2.5 ps of jitter required for NRZ data. Since the RZ pulse takes up a finite fraction of the bit window, RZ data should be capable of handling peak maximum optical input powers higher than those for NRZ data. Typically, RZ data systems should be able to handle +13-dBm peak power, whereas NRZ systems should be able to handle 3 dBm less, or +10-dBm peak power.

These specifications present a formidable challenge for the receiver designer to combine a PIN diode and amplifier without optical amplification. Currently, the easiest approach for RZ-formatted systems is an optical amplifier plus PIN diode; however, this means the PIN diode must handle a high peak optical power in order to drive the clock- and data-recovery circuits directly without benefit of electrical amplifiers. Non-return-to-zero-formatted systems, being less stringent in terms of bandwidth, can combine an optical amplifier, PIN diode, and electrical amplifier for their implementation. Both receiver architectures should be capable of producing from 250 to 500 mV of output drive voltage for the clock- and data-recovery circuitry.

Importance of the eye diagram
The eye-diagram pattern, which is the convolution of the transmitter and receiver response, provides the essential diagnostic tool for gauging the quality of optically driven transmission systems. A high-quality RZ eye diagram, using the optical amplifier plus PIN diode architecture, has several features to recommend it (see Fig. 2). The signal returns to baseline between successive pulses, demonstrating sufficiency of the system bandwidth. The jitter, based upon trace width at half-maximum, is low. The undershoot of the "1" bit is minimal, implying a smooth rolloff on the radio-frequency response and increased system sensitivity. (Overshoot and undershoot refer to time-domain signals as observed on an oscilloscope, where 0 and 1 bits are fixed voltage values. Undershoot is the dip in the observed waveform below the 0-bit voltage value on a transition from 1 to 0.) The device is direct-current-coupled, implying that the system sensitivity and eye diagram clarity is word-length independent. Finally, as a result of polarization-dependent loss in the detection process, the voltage variation of the "1" bit is minimal, which also helps lead to a clean and open eye diagram.

A high-quality NRZ eye diagram, using the optical amplifier plus PIN diode, plus electrical amplifier architecture, also has several features to recommend it. The center portion of the eye is clear; the jitter, as in the RZ case, is low; and both the overshoot and undershoot of the "1" bit are small, again implying a smooth rolloff and enhanced system sensitivity. In addition, the small low-frequency cutoff, as evidenced by the clean eye diagram and low bit-error-rate values for the long word length, adds to the system robustness. Furthermore, low PDL leads to small voltage variation in the "1" bit and to a clean eye.

As we have discussed, low polarization-dependent loss is a critical benchmark in measuring the efficacy of the O/E converter. Maximum values of 0.12-dB PDL for both RZ and NRZ systems help to ensure clean eye diagrams.

Optical amplifiers do not have the gain-bandwidth limitations of electrical amplifiers. Future 40-Gbit/s systems will use a combination of both optical and electrical amplification. As a result, optical-to-electrical converters will need high-bandwidth, high-power-handling capability, low PDL, and low ringing in order to meet 40-Gbit/s link requirements.

Roy Howard is a sales and applications engineer and Don Becker is an optical engineer at Discovery Semiconductors, 119 Silvia St., Ewing, NJ 08628. Roy Howard can be reached at rhoward@chipsat.com.

More in Home