With the introduction of 100G coherent interfaces in 2010, optical networking vendors began incorporating significantly stronger and more advanced forward error correction (FEC) algorithms into coherent DSPs and transponders. Today, modern 200G – 800G transponders operate so close to the theoretical Shannon Limit that even incremental improvements of a few tenths to a half-dB are viewed as significant achievements. Meanwhile, modern FEC algorithms provide a whopping 10- to 12-dB performance gain, easily the largest single impact on optical networking performance.
In communications networks, noise, nonlinearity impairments, and timing distortions all cause receive bit errors, as shown in Figure 1. FEC algorithms operate by encoding additional FEC, or parity, bits at the transmit (Tx) end that are used by the receive (Rcv) end to locate and correct any incoming bit errors.
Until recently, each optical networking vendor implemented their own proprietary FEC algorithms. While each vendor had their own “secret sauce” FEC implementations, almost all industry FECs operate utilizing a small number of underlying FEC codes. As a result, industry FEC performance gains tend to be very similar, especially when comparing FEC performance across the same generation of coherent DSPs. While proprietary FEC implementations continue to offer the highest performance, they come at the expense of multi-vendor interoperability. With the introduction of industry-standard 400ZR, OpenZR+, and Open ROADM interfaces, the industry has been slowly transitioning to common FEC algorithms to enable true multi-vendor interoperability, at least between coherent pluggable modules.
Understanding FEC performance
FECs are evaluated based on three key performance metrics:
- Net coding gain
- Overhead rate
- Pre-FEC BER threshold.
Net coding gain (NCG) is the optical signal-to-noise ratio (OSNR) performance improvement provided by a FEC-encoded signal compared to an uncoded wavelength. Modern high-performance FEC algorithms typically provide 10- to 12-dB NCG, which are astonishingly good performance gains.
Overhead rate, or redundancy ratio, is the ratio of FEC bits to information (data) bits. Up to a certain practical limit, the more overhead bits allocated for FEC processing, the higher the NCG and overall network performance. Most recent, high-performance FECs are designed using 15-25% overhead.
Finally, pre-FEC BER threshold is the worst-case incoming bit error rate (i.e., BER threshold), where the FEC algorithm still operates properly and provides (nearly) error-free communications after FEC decoding at the receive end (i.e., post-FEC BER). Communications systems never operate perfectly “error-free,” so the pre-FEC threshold is specified at a very low, nearly error-free post-FEC bit error rate, typically 10-15 BER. The benefit of modern FECs is they take really lousy, corrupted incoming signals and magically fix and transform them into (nearly) error-free output signals.
GFEC – The original
Reed-Solomon RS(255, 239), commonly known as “GFEC,” was one of the earliest optical FEC algorithms. GFEC was adopted as part of the ITU G.709 specifications and included as part of the standard OTN frame structure. As a result of the G.709 OTN standardization, GFEC became widely popular for 10G optical interfaces.
GFEC encodes 16 bytes of FEC data for every 239 data bytes, resulting in the RS(255, 239) naming structure. GFEC utilizes approximately 6% overhead for FEC bits and provides 6.2-dB NCG, as shown in Figure 2. By modern standards, GFEC is considered low to moderate strength. While GFEC worked well for 10G signals, newer, stronger, higher-performance FECs were needed with the introduction 100G+ optics.
Current common industry FEC types
Today’s FECs are usually based on one of three underlying FEC code types (Table 1):
- Concatenated FECs
- Block turbo codes
- LDPC codes
Concatenated FECs, also known as cascaded FECs or CFECs, combine inner and outer FEC codes, producing significantly improved performance compared to the original GFEC. A CFEC that combines a hard-decision staircase FEC (SC-FEC) outer code and soft-decision Hamming (SD-FEC) inner code was adopted as part of the OIF 400ZR Implementation Agreement for use on 400ZR coherent modules. CFECs utilize approximately 15% overhead and provide ~10.8-dB NCG. In addition to their use on 400ZR optics, CFECs were also commonly used on first-generation 100G optics.
Providing slightly higher performance, a block turbo code FEC algorithm has been adopted by the OpenZR+ and Open ROADM multi-source agreements (MSAs); this iteration is widely referred to simply as “oFEC” (for “Open FEC”). oFEC supports approximately ~11-dB NCG.
At the very high-performance end of the spectrum, vendors rely on proprietary FEC implementations based on low density parity check (LDPC) codes. While the underlying mathematics for LDPC codes were devised in the 1960s, it wasn’t until the 1990s that the algorithms were revisited for use in modern communications systems. And it was not until ~2015 when LPDC codes began to be widely implemented on commercial optical transponders. LPDC-based FEC codes typically provide 11.5- to 12-dB NCG.
Soft-decision decoding
Soft-decision decoding is another FEC technique to improve NCG performance. Traditional hard-decision FEC communication systems compared an incoming bit to a fixed threshold level. Anything that fell above this hard threshold is defined as a “1” bit and anything below the threshold is set to a “0” bit.
A soft-decision decoder provides a finer, more granular indication on whether the incoming signal really is a “1” or a “0” bit. Soft-decision decoders use additional soft-decision bits, or “confidence” bits, to indicate how far above or below a received bit is from a threshold. Due to the probabilistic nature of communication systems, bits that sometimes fall very near a threshold, either slightly above or below, can be misinterpreted and defined incorrectly. Soft-decision decoders, especially when coupled with iterative decoding, ensure those marginal bits falling very near a threshold are accurately received, as shown in Figure 3.
As just mentioned, soft-decision FEC decoding is typically coupled with “iterative decoding,” passing the received signal through multiple SD-FEC decoding stages within the coherent DSP. Each pass through a FEC decoder stage provides small, incremental performance improvements, again up to a practical limit. Almost all of the NCG benefit is captured by using three-stage iterative decoding; there is very little benefit to adding additional stages or passes.
The drawback to iterative FEC decoding is higher complexity and higher DSP power consumption. Due to the higher power consumption, iterative-stage FEC decoding is typically limited to a single stage in smaller pluggable modules, such as QSFP-DD modules.
Power of modern FEC
Modern FEC codes provide an astonishing 10- to 12-dB performance improvement and easily have the single biggest impact on transponder and optical network performance. Modern FEC codes are so strong that they operate to within fractions of a decibel of their theoretical coding performance.
Vendors traditionally implemented proprietary FECs and continue to base their higher performance transponder on proprietary LPDC-based FEC algorithms. Recently, the optical networking industry has transitioned to using common FECs for coherent pluggable optics, including CFEC used on 400ZR optics and oFEC used on OpensZR+/Open ROADM compliant optics. When coupled with soft-decision, iterative decoding, each of these modern FECs provide significant performance gains and are a key underlying technology that enables 400G+ coherent optics over metro to long-haul/subsea networks.
Special acknowledgement to Steve Jia, PhD for his excellent FEC article “Forward Error Correction (FEC): A Primer on the Essential Element for Optical Transmission Interoperability,” April 2019.
Randy Eisenach is part of the WDM and High Speed Optics Product Management team at Nokia. He specializes in optical transport technologies, next generation ROADM architectures, and high-speed photonics.