Adaptive approach needed for 10-Gbit/s backplanes

Digital signal processing technologies help ease these problems by allowing designers to craft adaptive architectures. By using an adaptive approach, designers can outfit their line cards with an ability to "tune" the various link parameters without expensive test equipment during development.

Designers can also craft systems that allow parameters to be monitored by system software on a real-time basis. That way continuous adjustments can be made when channel characteristics change. That's particularly useful when new cards are being plugged into the system, since each slot has different trace lengths and geometry on the backplane. A new generation of transceivers are now making this adaptive approach possible.

For years backplane-based systems evolved by moving to wider buses and faster signal clock rates. When 1 Gbit/s speeds entered the backplane realm, the reliable passing of data over parallel buses became impossible as signal skew and load problems increased. Designers were forced to shift from parallel buses to serial interconnects. Using serializer-deserializer (serdes) silicon, backplanes could deal with a serial stream that combines data and clock in the same signal.

While that strategy has been successful at the 1 Gbit/s level, new problems arise at faster speeds. Even at the 1 Gbit/s speed level backplane designers had to compensate for the problem of signal degradation over long traces. Dielectric losses and differences in impedance between the transmitter, transmission line and receiver typically account for this signal degradation. One technique called pre-emphasis boosts the higher frequency components of the signal. Higher-frequencies tend to attenuate faster than lower frequency components. The idea is to boost, right from the transmitter end, the high-frequency parts of a signal, knowing that it will attenuate by the time it reaches the receiver.

Pre-emphasis has become a mature technology, and most of today's 10 Gbit/s serdes chips employ it. Selective attenuation becomes an ever-greater problem at 10 Gbit/s and pre-emphasis is considered indispensable at such speeds. While some argue that pre-emphasis alone is a sufficient solution, there's a set of other problems which it doesn't solve at these higher speeds. In fact, using pre-emphasis as the sole solution these days is analogous to using only the volume knob to adjust the sound quality of a stereo, ignoring bass, treble, balance and so forth. Unfortunately most transceivers only offer the pre-emphasis "knob" as a means to tune signal quality.

Unfortunately, pre-emphasis at the transmitter alone does not effectively tackle the problem of InterSymbol Interference (ISI), the interference caused by adjacent symbols. Minimizing ISI also requires equalization at the receiver. At faster speeds ISI becomes a more severe problem. At slower speeds pulse widths are long, and you can sample far from the signal boundaries where symbols overlap. Approaching 10 Gbits/s, ISI is no longer a boundary case and it starts to affect the whole width of the bit. Using equalization, the transceivers can be tuned to minimize the effects of ISI. The key benefit of receive equalization is that it improves receiver performance without increasing the peak transmitter power.

Reflections caused by impedance mismatches also hurt performance. These impedance mismatches occur between the various connectors in a backplane. Reflections are often most severe in very short traces, where shorter transmission paths do not attenuate reflections, causing them to sometimes go back and forth many times before they dampen. An ability to adjust the input and output impedance of the transmitter termination is required to reduce such reflections.

Another factor impacting signal quality in 10 Gbit/s backplanes is crosstalk, the noise related to the coupling of signals on parallel transmission lines at the near end or far-end of the line or trace. Known as Near-End and Far-End Crosstalk (NEXT and FEXT), these impairments are dependent upon signal amplitudes, signal spectrum, and trace/cable length. Higher bandwidths compound these challenges, increasing the error rate of the transmission. An ability to do amplitude control helps minimize the effects of FEXT and NEXT noise.

While most 10 Gbit/s transceivers offer an ability to adjust only the pre-emphasis parameter, components such as Mysticom's MY3004 10G transceiver allow for a wider set of adjustments. On the chip's transmitter side the device supports adjustments to pre-emphasis, amplitude, and impedance. And on its receiver side it supports tweaking of equalization and impedance. By adjusting these five "knobs" system designers can better achieve the optimal level of signal quality for each backplane slot and card.

Parameters such as pre-emphasis must be tuned for specific slots in a backplane, and for the board that goes in each slot. As a result, in typical systems, backplane transceivers are tuned once, using oscilloscopes and pattern generators to determine optimum parameters. After extensive testing with a scope, systems designers come up with a table of values to give to the software team that is writing the drivers for the system. The table provides a pre-emphasis setting for each slot.

While that process works well for the initial deployed system it's not practical to go back and re-tune the transceivers once the system is deployed. As a result, any dramatic change - such as when a line or switch card fails, is replaced, or upgraded - leaves no easy way to get the same optimal signal quality enjoyed when the system was first deployed. That sort of "statically" tuned system works fine at lower speeds. But as backplanes approach 10 Gbit/s any change in hardware configuration throws those initial parameters out of whack, and the result can be huge variations in performance.

Next-generation designs need to shift to an adaptable backplane system approach. Adaptable programmable transceiver architectures are an important part of that approach. Such architectures enable designers to constantly adapt the transceiver to diverse conditions during development and in real-time. Designers can therefore significantly improve data transmission accuracy, reduce design complexity and maintain signal quality throughout the life of the equipment.

At higher speeds, the ability to tweak values on the fly is critical. Along those lines, Mysticom offers the hooks within its MY3004 10G transceiver to present a Signal Quality Index for software to access. The device aggregates several different signal quality parameters and aggregates them into a score. That score isn't just a simple bit-error rate, but rather a composite index that rates the overall signal cleanliness.

The score is accessible via a register on the device, from which software can read the score on the fly. That combined with the ability to adjust via software the five parameters -- pre-emphasis, amplitude, and impedance on the transmitter, plus equalization and impedance on the receiver -- results in an architecture that can retune itself while deployed. In addition, a 'C-like' pseudo-code can ease the task of integrating the adaptive software algorithms that make these features work.

With such an adaptive architecture, initial development is much easier, reducing the need for laborious testing with oscilloscopes. However, there is still a value in starting with a static table of values for each system slot - perhaps using the adaptive algorithms to determine the optimal parameters, and using scopes to verify them. The key is the ability to adjust the parameters in real-time once the system is deployed. For example, if someone pulls out an OC-192 line card from slot five in the backplane, and replaces it with a 10- Gbit Ethernet card, the adaptive system software would re-evaluate and tweak the parameters to get a new optimal setting.

The adaptive backplane approach has an immediate benefit for gross system changes in hardware -- when line cards or switch cards fail, or are swapped out, or moved to different slots in the backplane. In those cases, the need to retune transceiver settings is vital. Yet, there may be other more subtle opportunities where real-time adjustments could boost performance such as in the optimal settings for a system backplane that's not transferring any data compared to one that's running full bandwidth traffic on all cards could be significant. Adaptive architectures could handle that level of adjustments as well, and perhaps enhance system performance by doing so.

See related chart