>"Serial connections can be clocked over 100 GHz, and some folks are looking higher than that. “People are talking about 200GHz,” said Wendy Wu, director of marketing in the IP Group at Cadence. “But the use case for 200GHz isn’t clear yet.” To eliminate clock/data skew,
the clock isn’t forwarded to the receiver directly. Instead, it’s encoded onto the data stream
and recovered in the receiver by a clock-data recovery (CDR) block. That CDR block contains phase-adjustment elements like phase-locked loops (PLL), which also can be used for de-skewing purposes.
For that PLL to maintain lock, it must see sufficient transitions to keep it running.
A data stream having a long run of zeros would cause the PLL to lose lock. To avoid this, encoding schemes have been devised to ensure sufficient transition density. Because typical serial lines are AC coupled, these schemes also ensure that the number of 1s and 0s (so-called “disparity”) is balanced to ensure that the voltage on the lines remains centered." [...]
"A very different format: Chord
Another format, called Chord, was developed by Kandou to address this same space. Kandou describes it as a
generalization of two-wire differential signaling to n wires.
It’s much harder to understand than PAM-4 is, but the company claims greater noise immunity than PAM-4 provides. Reducing inter-symbol interference (ISI) in transmission lines involves a decision feedback equalizer (DFE), which cancels out the effects of prior bits on subsequent bits. The idea is to boost the data signal without also amplifying the noise. [...] The noise immunity is boosted by encoding data across the lanes. The data is multiplied by a collection of mutually orthogonal code vectors in the transmitter, and those are mixed onto the wires. At the receiver, the recovered signals are then again multiplied by the vectors – at least in concept – to recover the original bit streams. The size of the code vector will be equal to the number of wires carrying the signals.
This leverages a concept called “spatial orthogonality,” and it’s very similar to concepts used in code-division multiple access (CDMA) encoding for sharing a series of cellular frequencies between many different connections. It’s also similar to the codes applied to signals being transmitted and received through multiple-in/multiple-out (MIMO) and other beamforming systems.
The conditions for such orthogonality are that the sum of the vector entries must be zero to combat common-mode noise, and that the inner product of all of the code vectors must be zero for orthogonality."
It's probably something to hold us consumers over until we get optical connections everywhere... but, that being said, there's a whole bunch of really interesting Electrical Engineering concepts here!
>"Serial connections can be clocked over 100 GHz, and some folks are looking higher than that. “People are talking about 200GHz,” said Wendy Wu, director of marketing in the IP Group at Cadence. “But the use case for 200GHz isn’t clear yet.” To eliminate clock/data skew,
the clock isn’t forwarded to the receiver directly. Instead, it’s encoded onto the data stream
and recovered in the receiver by a clock-data recovery (CDR) block. That CDR block contains phase-adjustment elements like phase-locked loops (PLL), which also can be used for de-skewing purposes.
For that PLL to maintain lock, it must see sufficient transitions to keep it running.
A data stream having a long run of zeros would cause the PLL to lose lock. To avoid this, encoding schemes have been devised to ensure sufficient transition density. Because typical serial lines are AC coupled, these schemes also ensure that the number of 1s and 0s (so-called “disparity”) is balanced to ensure that the voltage on the lines remains centered." [...]
"A very different format: Chord
Another format, called Chord, was developed by Kandou to address this same space. Kandou describes it as a
generalization of two-wire differential signaling to n wires.
It’s much harder to understand than PAM-4 is, but the company claims greater noise immunity than PAM-4 provides. Reducing inter-symbol interference (ISI) in transmission lines involves a decision feedback equalizer (DFE), which cancels out the effects of prior bits on subsequent bits. The idea is to boost the data signal without also amplifying the noise. [...] The noise immunity is boosted by encoding data across the lanes. The data is multiplied by a collection of mutually orthogonal code vectors in the transmitter, and those are mixed onto the wires. At the receiver, the recovered signals are then again multiplied by the vectors – at least in concept – to recover the original bit streams. The size of the code vector will be equal to the number of wires carrying the signals.
This leverages a concept called “spatial orthogonality,” and it’s very similar to concepts used in code-division multiple access (CDMA) encoding for sharing a series of cellular frequencies between many different connections. It’s also similar to the codes applied to signals being transmitted and received through multiple-in/multiple-out (MIMO) and other beamforming systems.
The conditions for such orthogonality are that the sum of the vector entries must be zero to combat common-mode noise, and that the inner product of all of the code vectors must be zero for orthogonality."
It's probably something to hold us consumers over until we get optical connections everywhere... but, that being said, there's a whole bunch of really interesting Electrical Engineering concepts here!
Related: https://www.kandou.ai/