Without question, coherent detection has revolutionized optical networking and ushered in the age of 100G long-haul DWDM systems -- transmitting 100Gbit/s data rates over distances of thousands of kilometers, all on a single wavelength. Over the past couple of years, coherent DWDM systems have also moved "down market" into regional and metro-area networks.
But rapid growth, coupled with the unique demands of metro data center interconnect (DCI) applications, is challenging the dominance of coherent detection in DCI -- and particularly in DCI connections with spans less than 80km. Here, direct detection modulation has re-emerged as a strong potential contender for metro access DCI, and, specifically, the PAM-4 variant of direct detection.
Microsoft emerged as the champion of direct detect metro DCI when, at OFC 2016, it announced its partnership with Inphi for 100Gbit/s transceivers that use PAM-4 silicon, consume 4.5 watts of power, transmit up to 80km, and plug directly into data center switches (eliminating the need for external DWDM boxes for interconnection). Microsoft has said that 40km is the real "sweet spot" for its DCI applications -- a reach readily achievable with PAM-4 direct detect modulation. At OFC 2017, Microsoft, ADVA and Inphi published a technical paper detailing results of a 4Tbit/s commercial system and line successfully delivering 4 Tbit/s of capacity over an 80km link.
To be clear, direct detection is not going to eliminate the need for coherent detection -- even in metro DCI applications. But direct detect is rapidly emerging as an economically viable option in metro DCI. Still, there are significant trade-offs in choosing direct detect over coherent detection. We discuss the major puts and takes of PAM-4 compared to coherent detection below:
- Power consumption: PAM-4 eliminates the power-hungry DSP from the card/transceiver design, and therefore consumes significantly less power than coherent detect systems. As stated, the Inphi modules consume 4.5 watts per 100 Gbit/s, compared to about 50 watts per 100 Gbit/s for the best commercially shipping coherent detect modules today and 20 watts per 100 Gbit/s for the very latest announced systems.
- Size: The QSFP-based PAM-4 modules fit directly into client ports on a switch/router, meaning there is no capacity trade-off in switching from client optics to long-reach optics. Also, the QSFP28 is more than 80% smaller than a CFP2 module (by volume).
- Cost: Removing the DSP from the design removes a lot of cost from the modules. PAM-4 modules are cheaper than DP-QPSK and 16-QAM modulation formats.
- Distance: This is one major trade-off area in which coherent detection notches a win. DP-QPSK transmission distances are in the thousands of kilometers, and deployed all over the world. Alternatively, suppliers can use 16-QAM modulation to boost capacity per channel to 200 Gbit/s. 16-QAM distances are shorter than QPSK, but several hundred kilometers is commercially achievable today. On the other hand, direct detect PAM-4's distance is advertised at 80km maximum, and dispersion compensation is required to achieve anything more than several kilometers, as we discuss further below.
- Optical line complexity: Direct detect is more complex in deployment and maintenance compared to PAM-4. Coherent detection is far more tolerant of line impairments (due to the DSP) and, therefore, optical line system requirements are much simpler.
When 40G interfaces were introduced in the mid-2000s, chromatic dispersion (CD) and polarization mode dispersion (PMD) became big industry concerns and discussion topics. Chromatic dispersion is caused by different wavelengths (colors) in a pulse of light traveling at different speeds. PMD is caused by differences in propagation velocities for different polarization states. At sub-10G rates, CD and PMD tolerances are very high, but at 10G and above dispersion becomes significant. At 100 Gbit/s, CD and PMD was so high that long-haul distances were not achievable -- thus ushering in today's coherent era.
With the return to direct detect 100G (now focused on metro distances), CD and PMD must once again be measured. Dispersion-compensating fibers (DCFs) will be needed on the line for PAM-4 links greater than 5-8km, based on the published specs from existing PAM-4 modules. Fiber CD values must be entered into DCF modules for accurate performance, and this CD data comes from testing.
While dispersion compensation modules mitigate chromatic dispersion, they do not address the other major dispersion effect -- PMD. Test manufacturer EXFO recommends testing PMD for any spans longer than 10km -- which will be most links in the case of metro DCI applications targeted with PAM-4. PMD is measured with a PMD tester.
One simple but important consideration for dispersion testing is the use of single-ended testers versus dual-ended testers. A dual-ended tester requires a source at one end of a connection and a detector at the far end. This means that two technicians are required to perform the CD/PMD test. Newer single-ended testers combine the source and the detector within a single instrument, thus eliminating the need for a second technician (and thus saving on opex).
Beyond CD and PMD for PAM-4, basic "best practice" fiber test practices still apply. Fiber characterization tests with inspection probes and optical time domain reflectometers (OTDR) are recommended to identify causes of high attenuation, such as dirty connectors, fiber bends and poor splices. Finally, optical signal to noise ratio (OSNR) measurements are necessary, particularly for the longer PAM-4 links that will need amplification.
— Sterling Perrin, Principal Analyst, Heavy Reading
This blog is sponsored by EXFO.