The Great Gigabit Backplane Shootout - Question
#8
From a signal processing point of view, there is no difference in SXR (signal to crosstalk ratio) delivered to the receiver when the equalization is moved from transmit to receive, or anywhere in between. While some have claimed power savings based on receive equalization (transmitting smaller amounts of HF energy), simply transmitting a smaller signal with full equalization would produce the same effect: a small signal at the receiver demanding low thermal and extraneous noise. Which makes the system more fragile and requires additional power consumption to make a better receiver.
While the optimum equalization depends very little upon the coding technique, binary signaling is somewhat more tolerant of mis-equalization. However, at higher data rates, when the loss in the channel is larger than that seen by 3.125Gb/s binary SERDES operating today, accurately chosen equalization is a necessity for either signaling technique.
At 3Gbps, transmit pre-emphasis is sufficient for most backplane environment. At higher data rates, transmit pre-emphasis will need to be used in combination with receive equalization techniques for the lowest power/area solution.
Depending on the application, either is a good solution. Pre-emphasis tends to work very well either alone or with receive slope equalization over short to intermediate links, so would find good matches with chip-to-module construction as might be found in and optical backplane. For longer links, receive equalization (without pre-emphasis) appears to be superior under most realistic jitter scenarios, due to the fact that pre-emphasis increases cross-talk. If adaptive equalization is desired, receive equalization is a better choice. Equalization requirements tend to get more complicated when moving to more complex signals, as more analog circuitry must be used and linearity is extremely important.
Technical merits aside, for interoperability, SerDes should implement both techniques.
Once again depending on the specific application transmit pre-emphasis can be preferred over receive equalization or vice-versa or both. In general, transmit pre-emphasis requires minimum amount of extra hardware and power, however it is not adaptive to the trace length and can only be programmed. The receive equalization can vary from a simple first order high pass filter to a sophisticated DSP based filter. The advantage of the receive equalizer is that it can be adaptive and correct the ISI for different lengths. The amount of hardware and power for receive equalization in general is higher than pre-emphasis, however a simple first order adaptive equalizer can be designed to be very small and low-power. If the line coding requires a DSP to recover the signal, it would make sense to implement the equalizer in the DSP, otherwise analog equalizer can be sufficient.
Historically, transmit equalization was the only realistic solution for handling high frequency signal attenuation due to the larger process geometries available. This was effective in XAUI and lower speed SERDES. The problem with transmit pre-emphasis is that the high frequency energy added to the signal leads to much higher levels of crosstalk in connectors. As the signaling rates climb above XAUI speeds transmit pre-emphasis actually starts doing more harm than good making it a poor choice. Even systems with less noise sensitivity due to their line coding such as NRZ (Non-Return to Zero) will suffer dramatically at 6.25G signaling rates when using transmit equalization.
Receive equalization although more complicated from a circuit perspective, can solve many of these issues. Additionally, minimizing the signaling rate also simplifies this design considerably. An optimal solution is one that does not place excessive restrictions on board design /layout and thus is one that offers the most robust performance. Since receive equalization can be implemented many ways, KeyEye has always kept in mind the main limitation in these systems, noise.
Linear receive equalization, which works well in coax cabling environments due to the low noise levels, will tend to emphasize noise as well as the targeted signal in the backplane. It also leaves the device susceptible to issues such as channel loss ripple. This will ultimately place undue design restrictions on the backplane designer. A DFE-based receive equalizer is needed to provide the optimal performance in this noise limited environment.
Transmit pre-emphasis is the most optimum equalization technique for most copper media due to the following advantages:
Transmit pre-emphasis or receiver equalization alone will not be sufficient. A specially crafted system using both approaches is needed to compensate each other.
In our experience, especially when driving longer trace lengths, or to just provide a higher level of margin, both Tx pre-emphasis and rx equalization are required. Programmability or just the ability to turn them on/off are a necessity.
We are big fans of a mix of Tx pre-emphasis and Rx fixed pole equalization. Which is best? A blend is the most general, and therefore best for a general purpose application.