Bell Labs breakthrough will raise capacity limits of optical networks
Researchers at Bell Labs, the research arm of Alcatel-Lucent, have created a prototype technology – a 6x6 real-time multiple-input-multiple-output (MIMO) optical transmission system – that could make petabit optical systems a reality one day.
With network data traffic growth at cumulative annual rates of up to 100 per cent, Bell Labs estimates that within about a decade there will be an acute need for commercial optical transport systems capable of handling petabit-per-second capacities on a single optical fibre. Spatial division multiplexing (SDM) could provide a way to multiply the capacity of optical systems up to this level by exploiting parallel spatial channels, such as multiple fibre cores or modes.
Using this technique, the capacity ceiling imposed by Shannon’s theorem – which imposes a fundamental threshold for the maximum information transfer rate on a transmission channel – can be increased proportionally to the number of spatial modes. This effectively sidesteps the physical limit that applies to conventional fibre-optic systems built from standard singlemode fibre.
Laboratory demonstrations have already shown that petabit capacities can be achieved using special fibres in combination with SDM, but Bell Labs’ work takes things an important step further. The researchers have built a signal processing board that can unravel the interactions between different spatial paths in real time. Previous experiments have used sampling oscilloscopes and computers to perform the processing after the fact – clearly not a viable option in a real-world system.
‘We don’t resort to a PC for all the digital signal processing but we do this in real time on an FPGA [field-programmable gate array] platform, so this is the real breakthrough,’ said Peter Winzer, head of the optical transmission systems and networks research department at Bell Labs. ‘It’s still of course a research experiment, it runs on a research platform, but it is in real-time. But by implementing things on an FPGA, you take the first step towards real-world ASIC implementation.’
Details of the breakthrough were revealed at the 2015 IEEE Photonics Conference in October (download the paper here). In the experiment carried out at Bell Labs’ headquarters in New Jersey, coherent signals were transmitted and received over a 60-km-long coupled-mode fibre supporting six spatial and polarisation modes. The crosstalk between the different optical paths was successfully removed for the first time using real-time 6x6 MIMO digital signal processing (DSP).
MIMO techniques have been well studied in other areas of telecoms. In the wireless world, MIMO processing already been implemented commercially, and this technique is also exploited to remove crosstalk in VDSL vectored and G.fast systems over twisted-pair copper cables. MIMO’s application to optical networks is relatively new, but the same basic principles apply, Winzer says.
‘Polarisation multiplexing [in coherent optics] is just the simplest form of MIMO, it’s a two by two system,’ he explained. ‘You launch in x and y polarisations and then the polarisations rotate totally randomly in the fibre giving you some random crosstalk between x and y polarisation at the end of your fibre, and that is being processed out by the coherent chips that are on the market today.’
With three modes in two polarisations, the 6x6 MIMO-SDM system developed at Bell Labs is three times more complex than a singlemode fibre with two polarisations in today’s coherent optical systems. To reach petabit capacities would require the MIMO to be scaled up to support perhaps 10 or even 20 spatial channels. Luckily, scaling up the processing to more spatial channels is easier than it might first appear, as Winzer explained.
‘The MIMO processing engine in today’s coherent DSP chips is only about 10 per cent of the entire ASIC complexity. The rest is chromatic dispersion compensation, forward error correction and all the housekeeping functions. If you were to multiply [the MIMO part of the DSP chip] by a factor of 10, you end up with 100 per cent, which means that the entire ASIC almost doubles in size. So with only doubling in complexity of the entire chip, you would get 10 times more capacity through that chip.’
In order to be able to implement the MIMO DSP in real time, the Bell Labs researchers designed and built a complex 28-layer printed circuit board. ‘That design is absolutely not trivial,’ said Winzer. ‘It’s a very thick multilayer with thousands of traces feeding all this data into one FPGA to do the processing.’ The board differentially connects the coherent receiver outputs to twelve 5-giga-sample-per-second 10-bit analogue-to-digital converters (ADCs) made by e2v. These are interfaced to a single Xilinx XC7V2000T FPGA through 480 parallel lanes running at 1.25Gb/s each, resulting in an aggregate interface rate of 600Gb/s.
This FPGA board will now be used by the Bell Labs researchers to optimise the DSP MIMO algorithms – as the next step in developing future high-capacity optical systems that exploit SDM.
‘Studying channel dynamics would be the dedicated task of this real-time receiver,’ said Winzer. ‘With an oscilloscope the one thing you cannot do is monitor channel dynamics. You always take little snap shots of your data and then you process them. Being able over a long period of time to see how fast the channel varies, see how the various DSP algorithm react to those changes, that’s something you can only do with real-time hardware’.
In the meantime, the MIMO research could help Alcatel-Lucent devise better chips for its existing coherent optical systems. ‘We learned a lot about coherent algorithms while playing around with those MIMO algorithms in our labs,’ said Winzer. ‘MIMO is not going to be here in a commercial product across spatial paths within five years, but in preparing that future that’s five-plus years away, we at the same time understand the present much better and are able to implement more clever solutions into our present products.’