"NRZ is done. It's over. Move on," declared Ransom Stephens on January 30, the opening day of DesignCon 2018. For the 14th year, "The Case of the Closed Eye" panel convened on Santa Clara, CA. Following Stephens' declaration, the panelists — who represent IC communications companies and test & measurement — went on to explain that Stephens was partially right.

"Five years ago, PAM4 was a concept," said Chris Loberg of Textronix as he opened the panel. "PAM4 is being deployed everywhere except in short-reach links." Indeed, four-level pulse amplitude modulation (PAM4) has overtaken non-return-to-zero (NRZ, also called PAM2) modulation in all but the shortest distances over copper connections. PAM4 has become the standard, particularly in medium-haul and long-haul serial data links that run over fiber where each link transmits 50 to 56 Gbps per lane.

"NRZ is not dead," declared Intel's Mike Li. "At 56 Gbps, NRZ is still good for ultra-short reach applications. You could use PAM4, but it's expensive. NRZ circuits are much simpler. NRZ could work for extra-short reach (XSR) and very-short reach (VSR) applications."

Mike Li
Intel's Mike Li explains that PAM4 modulation has not taken over in short-reach data links.


Why is PAM4 winning? According to Li, it's because "we can't afford a different modulation to develop. We've solved 56 Gbps and now we're on to 112 Gbps."

PAM4 modulation A PAM4 signal has three eye openings.

Broadcom's Cathy Liu explained why PAM4 seems to be the modulations of choice, but that it has issues at 112 Gbps. "We ran experiments on PAM4, PAM8, and PAM16," she said. "At PAM8 and PAM16, the eyes are simply too small and thus PAM4 took over when the industry moved to 56 Gbps." At 112 Gbps, Liu said that the bandwidth of existing transmission channels will have to double because PAM8 and PAM16 aren't feasible. She may be right, but we can’t rule out the possibility of a breakthrough that will make them possible.

"PAM4 was made possible because of forward-error correction (FEC)," said Liu. That lets raw data arrive at relatively high data rates, say 10E-4 to 10E-7. With FEC, data rates can drop to 10E-12 and lower. "We still need to study pre-FEC data," she added. That's how to best characterize a transmission channel, with raw, uncorrected bits.

Inphi's Mark Marlett went a step further, saying that digital-signal processing will make 112 Gbps data rates possible. "DSP is enabling PAM4 optical channels," he said. "Reaches for 10 km, 40 km, and 80 km are enabled with DSP. We can also combine as many as 80 wavelengths on a fiber today. NRZ is dead in fiber. PAM4 and FEC are in."

Having established that PAM4 has won the battle in fiber-optic and other medium and long-reach applications, the panel turned to representatives of test & measurement companies to explain what needs to happen on that side.

DesignCon Closed Eye Panel
Martin Miller, Pavel Zivny, and Greg LeCheminant (l-r) listen prior to giving their talks on test and measurement at DesignCon's "Case of the Closed Eye" panel session on January 30, 2018.

Keysight's Greg LeCheminant looked at how we must treat errors with PAM4 because the bit rate is twice the baud rate. We need to look at symbol-error rate (SER) because PAM4 produces two bits per symbol. "SER is much lower with FEC but that doesn't help with test equipment designs or when troubleshooting. Errors can come in bursts, which can throw off overall SER measurement. It's no longer just about how many errors occur but how and where they occur. We have to account for long bursts of errors and errors across lanes."

Pavel Zivny of Tektronix brought up bandwidth in test equipment, something you'd think you can never have enough. There's now a feeling among engineers in the test community that it's possible to have too much bandwidth. Why? Because if your oscilloscope has more bandwidth than a receiver, it will add energy above the receiver's bandwidth to what you see on the screen. That's energy the receiver won't see, which could introduce errors into your measurements that might result in tests that produce false positives, showing no bit error where one might occur. "Oscilloscopes should have maybe just slightly more bandwidth than receivers," he said. Zivny also noted that standards call for oscilloscopes to use fourth-order Bessel-Thompson filters to limit bandwidth, but these filters have a long roll-off, which doesn’t match that of a receiver.

Teledyne LeCroy's Martin Miller wrapped up the panel presentations with a discussion of signal-to-noise and distortion ratio, a measurement that he questions. He noted that SNDR measurements are performed across an entire unit interval (UI) in an eye diagram. With PAM4, he claims only a measurement at the position of the widest eye opening is needed.

Martin Rowe covers test and measurement for EDN and EE Times. Contact him at [email protected]

Related articles: