[FAQ] DS320PR810: What happens if redrivers are cascaded?

We don't recommend cascading redrivers for a number of reasons: linearity problems, jitter accumulation, and difficulty of tuning.

Linearity problems:

• Redrivers have carefully designed equalization amplifiers that boost high-frequency components of incoming signals. The filters are shaped to compensate for the insertion loss of PCB traces and other media as best as possible.
  • "Linearity" is a major goal or feature of the equalization circuits. The idea is to successfully compensate for insertion loss (restoration of signal integrity) at various adjustable boost strengths without causing unwanted distortion of the signal in turn
• Because the actual loss profile of a system will deviate unpredictably from the ideal behavior of PCB loss (due to connectors, crosstalk, impedance changes, etc.), redrivers cannot perfectly compensate for signal degradation, so the redriver output will not be 100% identical to the original signal - there may be some "side effects" of a sort from redriver operation
  • However in a successful use case the restored shape will have good eye height and eye width, such that the receiver will have no problems accurately interpreting the signal, even if it's not the exact same as the original signal when looking at the frequency spectrum
• When two redrivers are cascaded, the transfer functions of multiple sets of non-ideal loss profiles and the two operating redrivers are all multiplied
  • The second redriver still sees some side effects of the first redriver's operation even after the signal has passed through the traces between the redrivers
  • Poles and zeroes remaining from the first redriver's filter can mix with the second redriver's filter and lead to distorted transfer functions - then there's a problem with maintaining linearity
• When linearity is poor, it's easy for incoming signals to be driven into a nonlinear range and suffer distortion
  • The final output seen by the receiver could have bad eye margins, or PCIe link training could be impeded by distortion of presets

Jitter accumulation:

• Jitter is a misalignment of periodic data signals such as a clock or a sequence of data
• Comes in different types, but the most basic one is "random jitter" (from imperfections in physical materials and circuits)
• Redrivers aren't designed to compensate for jitter and don't have any tools to compensate for it 
  • The best a redriver can do is to avoid amplifying what jitter is already present in the signal, and add as little new jitter from its own active circuitry as possible
• When a signal passes through a redriver, it will gain some amount of random jitter (~75 femtoseconds for DS320PR810) that should be outweighed by the signal quality improvements from the compensation of insertion loss
• With multiple redrivers in cascade, the jitter will add up or in bad cases even multiply, and this accumulation of jitter will be noticed by the receiver as an unpreventable reduction in eye width

Tuning difficulties:

• TI Gen 5 redrivers have 21 different EQ indexes for convenient device configuration, about half of which are commonly used
• With two redrivers, the number of possible EQ setting combinations becomes 441
• Even if there exists a perfect combination of cascaded redriver settings that will produce a result acceptable to the signal receiver (difficult due to nonlinearity problems), finding it may not be easy at all

Since we don't advise cascading redrivers, there are some alternatives when the performance of one redriver doesn't seem quite enough:

• Optimize the board layout to reduce sources of reflections, impedance chances, and crosstalk
  • Even if the signal path does not become any shorter as a result, these optimizations make it much easier for the redriver to successfully maintain signal quality across that distance
• Cascade a retimer with a redriver
  • Retimers have the ability to compensate for jitter, and have other digital tools for signal conditioning
  • Linearity problems are less likely, but still be sure to simulate and test thoroughly

Best,

Evan Su