Why Phase Interpolator Based CDR?

Поділитися
Вставка
  • Опубліковано 20 січ 2025

КОМЕНТАРІ • 24

  • @sunkarasaigoutham
    @sunkarasaigoutham 2 роки тому +2

    Thank you so much for such high quality videos

  • @dtfan4779
    @dtfan4779 Рік тому +1

    Hi Mr. chen,
    Thanks for sharing your knowledge,
    if you can, please guide me how to estimate PI-base CDR RJ on jitter budget?
    =>i think PI-baseCDR_Rj roughly estimate can be =(PLL_Rj^2+PI_Rj^2)^0.5 ,right?

    • @circuitimage
      @circuitimage  Рік тому

      Hi DT, you ballpark estimation of the total RJ is correct to me since both RJ are indepedent.

  • @rangababuganta7725
    @rangababuganta7725 2 роки тому

    Can u talk more about later ( 9.17/11.38 in video). about Latency !!
    Thank you Chen

    • @circuitimage
      @circuitimage  2 роки тому +1

      Yes, that was on the plan and thanks for the reminder.

  • @許育瑛-n5t
    @許育瑛-n5t 2 роки тому +1

    Thank you for your sharing about PI knowledge, I really appreciated it, and there are some questions I want to inquire to you.
    1.During the introduction of PI, you mentioned there are 8 phases, is it fixed or could it be more phases than 8
    2. In the video, you mentioned that every CDR loop may require a PLL to achieve the feature, so does it mean if there are three channels, then it needs three PLLs
    3. In the video, you also mentioned that Wi will select two input reference clocks, like CKin0 and CKin1. Afterwards, the selected clock(CKin0, CKin1) by Wi will go through the VCO and send it to PD again, so that we can get clock recovery. I'm not sure it's correct or not
    Thank you again for everything you've done

    • @circuitimage
      @circuitimage  2 роки тому

      Thanks for the feedback. I have comments below.
      1. Yes, you could have more than 8 phases as the PI's basis, but the complexity is too high in a multi-phase VCO. Both phase noise or frequency may not be feasible.
      2. Yes. If a PLL-based CDR, you may have three RX PLLs in three channels. See: ua-cam.com/video/ZJkadLjOg2s/v-deo.html
      3. No, the selected clock(CKin0, CKin1) by Wi will go through the PI's input to generate a finer phase of CKout. See slide #2 of ua-cam.com/video/FkoJwnAYbMs/v-deo.html

    • @許育瑛-n5t
      @許育瑛-n5t 2 роки тому

      @@circuitimage Thank you for replying, it's really helpful

    • @circuitimage
      @circuitimage  2 роки тому

      @@許育瑛-n5t I'm glad it helps :)

    • @許育瑛-n5t
      @許育瑛-n5t 2 роки тому

      @@circuitimageHi Mr.Chen, I still have some questions.
      1. Because PI can rotate at a rate equal to offset, PI can generate a fine phase of CKout right. But CKin0 and CKin1 are selected by Wi, seems like there is still a little variation range of phase for CKout, so if we increase more phases as input(like 15 or 20), it'll be more precise right, but it may let the circuit be complicated.
      2. Once Din inputs into PI system, it'll be generated a CKref by MP-PLL to PI, and CKref will be screened out two CKin by Wi, and then we can get the CKout, is it correct.
      Thank you very much.

    • @circuitimage
      @circuitimage  2 роки тому +1

      @@許育瑛-n5t Thanks for those good questions.
      1. You are correct. So, the Wi should be in the CDR loop. The Wi is dynamic changing as well to cover the offset.
      2. That's correct

  • @sunkarasaigoutham
    @sunkarasaigoutham 2 роки тому

    Hello Mr Chen,
    In one of the IPs we designed with PLL based CDR, we were able to tolerate +/-300ppm (and actually can tolerate 8x than this) of frequency offset between the TX and RX clock. How much range would a typical PI based CDR has?

    • @circuitimage
      @circuitimage  2 роки тому

      Good question. That depends on the PI's performance.

  • @narendrakundrapu2508
    @narendrakundrapu2508 Рік тому

    Tqs Chen

    • @circuitimage
      @circuitimage  Рік тому

      Hi Narendra, nice to meet you and thank you so much for your feedback.