Speedy InP chips look to boost optical network capacities

The burst of the internet bubble delivered a lasting impact on component suppliers to optical networks. Despite rocketing growth in internet traffic over the last few years, InP chip makers have struggled to make a profit in a market of falling prices. Instead, their focus has been on survival.

Recently though, things have been getting a little better. Pinch points in operators networks are prompting a transition from 10 Gbit/s systems to 40 Gbit/s versions. These employ faster components with advanced modulation schemes that should also form the foundations for the emerging 100 Gbit/s transmission market.

Operators are also considering 40 and 100 Gbit/s transmission for their own backbone networks. These firms estimate that core backbone internet protocol (IP) traffic is growing at 45–65% per year. Meanwhile, online video growth is driving data center networking requirements for the emerging 40 and 100 Gigabit Ethernet standards.

Faster transmission speeds can also deliver operational and management efficiencies. Aggregating 10 Gbit/s traffic into fewer, higher-capacity links simplifies an operator s management and equipment inventory. "It results in fewer [networking] boxes and fewer interfaces, and bigger pipes are good for operations," said Glenn Wellbrock, director of backbone network technology at Verizon Business.

The trigger for 40 Gbit/s occurred in 2005, when Cisco Systems and Juniper Networks used this for the interface on core IP routers. However, volume shipments of 40 Gbit/s transponders and systems had to wait until 2007, according to Daryl Inniss, vice- president and practice leader, communications components, at market research firm Ovum.

Inniss says that the 40 Gbit/s market can be divided into two. One sector is very-short-reach (2 km) client-side transponders that link core routers to dense wavelength division multiplexing (DWDM) equipment. The other is 40 Gbit/s line-side interfaces for long-haul transmission. Last year these sectors were worth $75 million, and they will generate $900 million by 2012. By then, line-side interfaces will have shifted from line cards based on discrete components to transponders.

Existing 40 Gbit/s interfaces are based on the SONET OC-768 data-rate standard. However, interface options will expand, thanks to the addition of 40 and 100 Gigabit Ethernet standards that should be completed in 2010.

High-speed designs
Implementing advanced modulation schemes delivering faster speeds is challenging because there are inherently fewer photons in the optical signal. On top of this, dispersion increases with speed.

However, three advanced modulation schemes capable of handling the extra speed have emerged: duobinary modulation; and the phased-based schemes of differential phased-shift keying (DPSK) and differential quadrature phased-shift keying (DQPSK) (table 1). These exist alongside the existing intensity-based non-return-to-zero (NRZ) modulation scheme used for very-short-reach 40 Gbit/s transponders.

Duobinary, the simplest of the three, involves the interference of data signals – "0s and 1s" are replaced with a three-level signal. Circuitry is relatively simple, but the technology is unsuitable for long-haul transmission, and operators such as Verizon Business favor phase-based modulation. DPSK components already exist, while DPSK transponders are appearing from JDSU, Opnext and Finisar. There is growing interest in the more advanced DQPSK. This is more tolerant to dispersion, which makes it a promising building block for 100 Gbit/s systems.

DPSK encodes data using the relative phase of consecutive bits, with zeros and ones coded as "no change" or "change" in phase. The optical signal is always on, unlike an NRZ-coded signal, and it s just the phase that changes between bits. Absolute phase doesn t need to be controlled – it is the difference between bits that matters, hence the "D" in DPSK.

The DPSK transmitters that have been produced all use a laser source and a modulator. The modulator can be a direct-phase version, but Mach Zehnder (MZ) intensity modulators are becoming popular. These are driven from the on state, through off and back to the next on state. Operated in this way, the MZ modulator becomes a very precise digital phase modulator, giving direct coding of PSK information onto the optical signal.

Recovery of the bit stream requires phase sensitivity from the receiver. The simplest way to implement this is with a delay line interferometer and balanced photodetectors. An incoming bit is delayed and interferes with the subsequent bit, which converts the phase information to intensity after the interferometer.

DQPSK extends the technique and four phases are used to represent the binary data. This halves the signalling rate compared with intensity-based, duobinary and DPSK modulations, which use 40 Gbit/s electronics. This means that DQPSK works with less demanding 20 Gbit/s electronics and the slower signalling suffers fewer dispersion impairments. It also works with 50 GHz DWDM channels, while DPSK needs to be adapted to meet such spacing.

However, these benefits come at the expense of the greater complexity of DQPSK s transmitter and receiver. A single MZ modulator arm is used to encode the DPSK signal, but two arms are needed for DQPSK. In effect, two parallel 20 Gbit/s systems decode DPSK streams.

Bookham is developing DQPSK modulators that are designed to be used with its tunable laser chips. "For 40 and 100 Gbit/s we wouldn t say we will never [monolithically] integrate [the two], but every decibel is important," explained Andy Carter, Bookham s vice-president of technology. He believes that a hybrid approach also speeds time to market. The 40 Gbit/s DQPSK encoder is targeting the cost-conscious metro/regional network (100–800 km), but it can also serve the long-haul sector.

One design challenge is the radio frequency – getting 26 GHz electrical signals in and out of the device. The modulator employs on-chip detectors. "The device needs a sophisticated set-up process, such as balancing the phase in the two [MZ] arms," said Carter. "The detectors enable this to be automated."

US system vendor and InP integration specialist Infinera is also backing DQPSK. This February it demonstrated an InP-based 10 × 40 Gbit/s DQPSK transmitter photonic integrated circuit (PIC) in the lab. This extends the capacity of its existing PIC from 100 to 400 Gbit/s, which it claimed is the most efficient way to reduce transport costs. Infinera is backing faster rates by halving the channel spacing and switching to DQPSK, which extends the capacity of each channel.

The DQPSK PIC swaps Infinera s existing electroabsorption (EA) modulator with a MZ in each of the array s 10 lanes. "The structure is the same as discrete implementations of a 40 Gbit/s DQPSK modulator," explained David Welch, Infinera s chief strategy officer. The number of components per channel is significant with DQPSK. This drives up the optical losses, but they are compensated by the introduction of distributed-feedback (DFB) lasers.

Infinera s latest system, the ILS2, uses its existing 10 × 10 Gbit/s PIC and has a passive PIC that enables 25 GHz channel spacing. "It will also use in future the DQPSK modulation PICs," said Welch.

The company is now working to improve its demonstrator PIC s performance, its sensing and control, and particular DQPSK receiver PIC challenges. "There is also the 400G electronics in and out, the packaging and the silicon ASIC (application specific IC) to feed the 400 Gbit/s phase-modulated signal," said Welch.

System and component vendors are also eyeing the emerging 100G Ethernet market. At the Optical Fiber Conference, Opnext presented a paper on an integrated InP 25 Gbit/s DFB and EA modulator.

Opnext s device is part of a coarse WDM (CWDM) 100 Gigabit Ethernet transceiver design to meet the IEEE s single-mode fiber 10 km reach specification. A CWDM design was selected because the wide wavelength spacing between the four channels provides a sufficient margin for the lasers to drift with temperature without needing a cooler. Avoiding using a thermoelectric cooler reduces transceiver power consumption by up to 3.5 W.

Getting the directly modulated DFB to work at 25 Gbit/s demands a reduction of the device s overall capacitance. "For a directly modulated laser, you want as short a [laser] cavity length as possible because it acts as a capacitor, reducing the overall frequency response," said Matt Traverso, senior manager of technical marketing at Opnext.

In a DFB/EA, the laser and modulator are isolated, which improves modulation at 25 Gbit/s. Although the 100 Gbit/s standard is not completed, Opnext believes that the DFB/EA can meet the final specification while delivering a high yield.

Any resulting Opnext transceiver product for 100 Gbit/s would use discrete packaging or hybrid integration – the firm has not disclosed its preferred approach. A discrete design would employ existing manufacturing techniques, but a hybrid design would have manufacturing benefits at high volumes.

In May the IEEE Task Force adopted closer spacing between channels – dubbed LAN-WDM – for the 4 × 25 Gbit/s specification. Some standard group members believe that the design will be easier to produce monolithically even though a thermoelectric cooler will be required. Opnext s CWDM DFB/EA does not meet the requirements, but it is now adapting its device for a LAN-WDM implementation.

Balancing detectors
U2t Photonics is one of several firms offering balanced detectors for DPSK and DQPSK. It is also working with partners to develop a balanced receiver that integrates the delay line interferometer with balanced detectors. This will simplify 40 Gbit/s transponder and line-card design.

Detectors for 10 Gbit/s requirements are surface illuminated. However, with this configuration the responsivity – the efficiency of converting the optical signal to an electric current – is traded with device bandwidth. This means that at 40 Gbit/s the surface-illuminated detector s bandwidth is limited unless responsivity is sacrificed. To overcome this, U2t s photodiode is edge-illuminated by a waveguide.

The detector s power linearity is also important, especially for long-haul applications that employ a preamplifier at the receiver. "Power linearity [of edge detection] is much improved," said Jens Fiedler, U2t s vice-president of sales and marketing, who revealed that absorption is well distributed across the volume of the detector.

Great care is needed to control the amplitude of the output in a balanced receiver so that it is equal in both detectors. Phase must also be balanced, so that the pair of detectors have the same phase dependencies and transit times.

The company also offers 100 Gbit/s detectors, which are initially for use in the lab and for test equipment. "Optimizing the design for 100 Gbit/s requires slight changes," said Fiedler, such as the detector s layer structure and the electrical circuits. U2t is also sampling its 100G Ethernet detectors. "We are awaiting the market pull and the final adjustments to meet customer needs," said Fiedler.

"The client side and the line side are challenging at 40 and 100 Gbit/s," commented Ovum s Inniss. "There are so many different solutions – serial and [wavelength] multiplexed."

As a result, optical component makers are facing hard decisions, but operators are providing an incentive. Verizon Business intends to adopt 100 Gbit/s transmission as early as 2009. "The target is 2009. Will we get it? That s debatable," said Wellbrock. "We certainly want 100G as soon as possible and we re pushing the industry in that direction."   

View pdf of article