• State TI Thinks Resolved
  • Locked Locked
  • Replies 4 replies
  • Answers 1 answer
  • Subscribers 33 subscribers
  • Views 180 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

DS320PR810: Equalization performance confirm

QIANWEN WANG
Prodigy 110 points
Part Number: DS320PR810


Tool/software:

Hi TI support,

I have a question regarding the equalization performance of the DS320PR810 redriver. We are currently operating at PCIe Gen4 (16 GT/s), and the channel topology is shown below for your reference.

The redriver settings are configured as follows:

Additionally, the Tx EQ for the Root Complex (SW) is configured with Preset P5.

However, when observing the eye diagram at the Endpoint (EP), we noticed that the signal appears distorted.

Could you please help review this case?
Do you think the distortion might be caused by over-equalization introduced by the redriver?

Any suggestions or insights would be greatly appreciated.

  • 0
    TI__Mastermind 18800 points

    Hi Qianwen,

    If I am understanding your loss profile correctly, the total link loss is 9.09 dB + 5.9 dB + 1.7 dB = 16.69 dB ≈ 17 dB, and this would be for PCIe Gen 4 (symbol frequency 16 GHz, Nyquist frequency 16/2 = 8 GHz). According to the PCIe Gen 4 specifications, the allowable link loss is 28 dB. If your system has much less the 28 dB loss budget, this would not be a situation where we would expect a redriver to be helpful because the PCIe TX and RX can already provide enough signal conditioning to maintain good eye margins. The redriver equalization is unnecessary, could be excessive, and adds an extra stage of random noise. Redrivers are intended for cases with more total loss than the PCIe loss budget - was there a specific reason why your designers were interested in using a redriver in this project? Correct me if I have understood anything incorrectly.

    The "Default" EQ setting in our Gen 5 redrivers is a medium amount of equalization, and in situation with very low loss it's possible that 6 dB boost would be over-equalizing the signal and causing distortion. You can experiment by setting the EQ settings to EQ Index 0, which is the minimum equalization setting with 3 dB boost at 8 GHz, and observe if the shape of the eye diagram becomes more normal.

    Best,

    Evan Su

  • 0 in reply to Evan Su
    Prodigy 110 points

    Hi Evan,

    Thanks for your support. 

    I'd like to correct the previously shared channel loss profile — the package loss of both the switch (SW) and the endpoint (EP) was not included. The EP package contributes approximately 2 dB of loss. Although the exact loss of the SW package is unknown, since it's a PCIe Gen5 switch, we assume it follows the typical Gen5 root complex guideline, which is approximately 5 dB at 8 GHz.

    With that considered, the total channel loss should be:
    17 dB (channel) + 5 dB (SW PKG) + 2 dB (EP PKG) = 24 dB

    This system was originally designed for Gen5 operation. Due to the high channel loss at 32 GT/s, the redriver settings were determined during the initial design phase and have not been adjusted for Gen4 operation.

    We’ll certainly try your suggested adjustments. That said, I have a follow-up question:
    As I understand, the redriver is primarily intended to compensate for pre-channel loss. In our case, the pre-channel loss is about 14 dB. Given that, why would a 6 dB CTLE gain cause over-equalization? I would appreciate any insights you can provide on this behavior.

    BR

    Qianwen Wang

  • 0 in reply to QIANWEN WANG
    TI__Mastermind 18800 points

    Hi Qianwen,

    QIANWEN WANG said:
    As I understand, the redriver is primarily intended to compensate for pre-channel loss. In our case, the pre-channel loss is about 14 dB. Given that, why would a 6 dB CTLE gain cause over-equalization?

    From my past experience with many customer projects, the interaction between insertion loss and EQ boost is not so clear that it could be precisely calculated that way. Although loss and boost are represented by magnitude at Nyquist frequency so we can have easily compared dB figures, that is a sort of simplification that does not account for the full frequency response curves of the loss profile and CTLE amplifiers. Because EQ boost cannot be a perfect compensation for a real system's insertion loss, the differences lead to non-ideal behavior, especially once other factors like the amplifier linearity and interactions with PCIe presets or receiver EQ are considered. So often it is necessary to manually test and adjust redriver EQ in a real system after the board is manufactured.

    In your case, it's possible that the EQ Index Default setting has either too much boost or a less-compatible filter shape for your system's unique conditions and then does not mix very well at the final receiver eye diagram even though it would seem that the 6 dB gain at 8 GHz would seem fine. But in the opposite trend I have seen many applications where a medium redriver setting (medium amount of EQ boost) leads to a longer reach extension (covers large amount of insertion loss) than expected. In those cases I have assumed that the redriver EQ happened to mix well.

    We have an application note on redriver equalization that you may find interesting, it covers most of the points I discussed here and some extra.

    Best,

    Evan Su

  • 0 in reply to Evan Su
    Prodigy 110 points

    Thank you, Evan. We'll proceed to adjust the EQ index and observe whether it improves the eye quality. We'll keep you updated as soon as we have any findings.