Nice video. I have heard about this crosstalk issue before, but was not aware the mechanism behind it. Thanks for explaining. One question on terminology: When we say that the CTLE equalizes 15db loss, do we mean that its peaking gain is 15dB? Can we think about this value as the difference between CTLE's gain at Nyquist frequency and CTLE's DC gain. For instance, lets say at Nyquist frequency gain is 5dB and DC gain is -10dB, then we have peaking gain = 5dB - (-10dB) = 15dB.
Hi Hrachya, thanks for the good questions. The peaking gain is not gain at the Nyquist rate, which confused people all the time; therefore, you're correct. Peaking gain is the value as the difference between CTLE's gain at Nyquist frequency and CTLE's DC gain. Thank you so much.
Very useful. When you check SBR on the slide 5, did you apply differential 1V for 1UI and ‘differential 0V’(it means common mode voltage for both INP and INN) for all other UI to the input? Thanks !
Thanks for the feedback. Correct. For the Single-bit response (SBR), I applied differential 1V for 1UI and ‘differential 0V’ (common-mode voltage for both INP and INN) for all other UI to the input.
Thanks for the feedback and I'll share more circuit images of the Rx detection, resistor calibration, slicer calibration, CDR loops and IQ calibration, etc later.
I am interested in the CTLE's peaking gain & BW tradeoffs. In your slide "Summary (2)" there is a graph that shows BW reduces as the number of stages increases. I am wondering if there is any analytical explanation as to why there is such a trade-off? Thanks for your comment in advance.
Thanks for your nice video! In slide 9, or at 4:45, you mentioned the blue curve H(s) is a simulation result, is there a fundamental reason why this response is high-pass? If I understand correctly, the red curve (Xtalk+CTLE) is just a summation of the blue one and the green one, right?
Hi Chen, thanks for the video. At 8:10, the takeaway is that after 30dB loss (without Tx FFE), the high frequency swing is so low that its difficult to distinguish from Noise/Crosstalk (poor SNR) and any CTLE boost would just produce a closed eye at its output? Please correct me if I am wrong.
Hi Yujian, nice to meet you and thank you so much for the feedback. The FFE can mitigate the impedance discontinuity since it's a discrete-time implementation. But, the TX FFE is not quite efficient in power if a longer post-cursor or taps (from the reflection) is needed. But the DFE could have a much longer post-cursor. and low power compared to the TX FFE. Please be advised that the number of taps depends on the round trip of the reflection. Thank you very much again for the good questions. 😀
@@yujianzhai Thank you so much for the feedback. Here ruclips.net/video/BVM3HMCVtuA/видео.html has some frequency domain analysis of the TXFFE. As you can see, more taps can provide a more highly programmable frequency response of the gain shaping, which maximizes the SNR by reducing the reflection frequency content? 😀
@@circuitimage thanks chen,i mean if there are some big difference between the frequency response of CTLE and FFE，jiust like the FFE‘s frequency response has notch?
@@yujianzhai Thank you so much for your clarifications. That's what I mean by that the peaking (gain & frequency) of a CTLE would be coupled or constrained together. But, the FFE with more taps more than 10 taps (presumably), would have more flexibility to adjust both different peakings in gain & frequency, separately. But, the overhead is to provide as many taps as we could. If we only have 1 or 2 taps, that might not be the case in the frequency response.
Hi, Thanks for the wonderful video! I was just curious as to why the differential mode of operation wouldn't be able to reject the crosstalk to a level that it would become insignificant even after CTLE amplification. Is it due to mismatch? Thank you!
Hi Somanshu, nice to meet you and thank you for the good question. Yes, part of the mismatch, but part of it could be the big interference to disturb the operating point of the transistor even the common mode. That's part of the difficulty of the CMSI in PCIe's spec.
Very good video, how crosstalk in CTLE is amplified was well explained. Could you please also explain why crosstalk in TX-FFE will be amplified too? thank you very much . CC
@Meilin Wu Thanks for the good question. Yes, both CTLE & TX-FFE are similar since both cannot differentiate the noise & signal. The DFE can since the correct decision has been made and feedback cancels the ISI, not the noise, which is the beauty of the nonlinear slicer.
@@circuitimage Hello CC, thank you for the response. I understand that CTLE amplify crosstalk because of the Positive Gain, but since the TX-FFE does not have a positive gain, could you please elaborate how it amplify the crosstalk? Thank you very much.
@@kellyrolla Thanks for the good question. You are right about the TX-FFE, which would only reduce or attenuate its DC gain (signal). That would be another model interpreting the signal-to-noise ratio is attenuated or the noise-to-signal is amplified.
Nice video. I have heard about this crosstalk issue before, but was not aware the mechanism behind it. Thanks for explaining.
One question on terminology: When we say that the CTLE equalizes 15db loss, do we mean that its peaking gain is 15dB? Can we think about this value as the difference between CTLE's gain at Nyquist frequency and CTLE's DC gain. For instance, lets say at Nyquist frequency gain is 5dB and DC gain is -10dB, then we have peaking gain = 5dB - (-10dB) = 15dB.
Hi Hrachya, thanks for the good questions. The peaking gain is not gain at the Nyquist rate, which confused people all the time; therefore, you're correct. Peaking gain is the value as the difference between CTLE's gain at Nyquist frequency and CTLE's DC gain. Thank you so much.
Very useful. When you check SBR on the slide 5, did you apply differential 1V for 1UI and ‘differential 0V’(it means common mode voltage for both INP and INN) for all other UI to the input? Thanks !
Thanks for the feedback. Correct. For the Single-bit response (SBR), I applied differential 1V for 1UI and ‘differential 0V’ (common-mode voltage for both INP and INN) for all other UI to the input.
Hi, Excellent videos. Please post more videos Rx detection, resistor calibration, slicer calibration, CDR loops and IQ calibration
Thanks for the feedback and I'll share more circuit images of the Rx detection, resistor calibration, slicer calibration, CDR loops and IQ calibration, etc later.
I am interested in the CTLE's peaking gain & BW tradeoffs. In your slide "Summary (2)" there is a graph that shows BW reduces as the number of stages increases. I am wondering if there is any analytical explanation as to why there is such a trade-off? Thanks for your comment in advance.
Thanks for the feedback. I'll make an analytical explanation of CTLE's peaking gain & BW tradeoffs video.
@@circuitimage can u share any paper to read about this tradeoff in the meanwhile.
@Zheng Lai Here you go: ruclips.net/video/jqo6BqKzdQg/видео.html
Thanks for your nice video! In slide 9, or at 4:45, you mentioned the blue curve H(s) is a simulation result, is there a fundamental reason why this response is high-pass? If I understand correctly, the red curve (Xtalk+CTLE) is just a summation of the blue one and the green one, right?
Thank you so much for the fundamental questions. The channel is low pass, so you need high pass equalization to equal the frequency response.
Thank you😊
Hi Lab_mangement, you're very welcome. :)
Hi Chen, thanks for the video. At 8:10, the takeaway is that after 30dB loss (without Tx FFE), the high frequency swing is so low that its difficult to distinguish from Noise/Crosstalk (poor SNR) and any CTLE boost would just produce a closed eye at its output? Please correct me if I am wrong.
Yes, you are correct about the takeaway. :) Thanks for your feedback again.
Awesome explanation.
Hi Lavanya, thank you for your feedback and I'm glad that helped. 😊
Thank you Chen..
Welcome 😊
Hi chen,I wonder why FFE can also solve the problem of impedance discontinuity,just because it is also the nonliner system as DFE?
Hi Yujian, nice to meet you and thank you so much for the feedback. The FFE can mitigate the impedance discontinuity since it's a discrete-time implementation. But, the TX FFE is not quite efficient in power if a longer post-cursor or taps (from the reflection) is needed. But the DFE could have a much longer post-cursor. and low power compared to the TX FFE. Please be advised that the number of taps depends on the round trip of the reflection. Thank you very much again for the good questions. 😀
i can only undestand from time domain analysis ，but not from frequency domain analysis。Can you give some analysis from frequency domain?
@@yujianzhai Thank you so much for the feedback. Here ruclips.net/video/BVM3HMCVtuA/видео.html has some frequency domain analysis of the TXFFE. As you can see, more taps can provide a more highly programmable frequency response of the gain shaping, which maximizes the SNR by reducing the reflection frequency content? 😀
@@circuitimage thanks chen,i mean if there are some big difference between the frequency response of CTLE and FFE，jiust like the FFE‘s frequency response has notch?
@@yujianzhai Thank you so much for your clarifications. That's what I mean by that the peaking (gain & frequency) of a CTLE would be coupled or constrained together. But, the FFE with more taps more than 10 taps (presumably), would have more flexibility to adjust both different peakings in gain & frequency, separately. But, the overhead is to provide as many taps as we could. If we only have 1 or 2 taps, that might not be the case in the frequency response.
Hi,
Thanks for the wonderful video! I was just curious as to why the differential mode of operation wouldn't be able to reject the crosstalk to a level that it would become insignificant even after CTLE amplification. Is it due to mismatch? Thank you!
Hi Somanshu, nice to meet you and thank you for the good question. Yes, part of the mismatch, but part of it could be the big interference to disturb the operating point of the transistor even the common mode. That's part of the difficulty of the CMSI in PCIe's spec.
@@circuitimage I see. Thanks a lot!
@@somanshumishra1961 Hi Somanshu, you're very welcome.
Very good video, how crosstalk in CTLE is amplified was well explained. Could you please also explain why crosstalk in TX-FFE will be amplified too? thank you very much . CC
@Meilin Wu Thanks for the good question. Yes, both CTLE & TX-FFE are similar since both cannot differentiate the noise & signal. The DFE can since the correct decision has been made and feedback cancels the ISI, not the noise, which is the beauty of the nonlinear slicer.
@@circuitimage Hello CC, thank you for the response. I understand that CTLE amplify crosstalk because of the Positive Gain, but since the TX-FFE does not have a positive gain, could you please elaborate how it amplify the crosstalk? Thank you very much.
@@kellyrolla Thanks for the good question. You are right about the TX-FFE, which would only reduce or attenuate its DC gain (signal). That would be another model interpreting the signal-to-noise ratio is attenuated or the noise-to-signal is amplified.
You might have already explained, but why cannot CTLE cancel the ISI due to long tail?
Thanks for the feedback. The CTLE can cancel the ISI due to the long tail. Am I misunderstanding your question?