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.

DAC38J84: Documentation Question: Do Manual Sections 7.3.2 and 7.3.3 Apply Consistent Meanings of "SerDes PLL Output Clock?"

Part Number: DAC38J84
Other Parts Discussed in Thread: LMK04828,

Hi,

My company is programming the DAC chip directly, without relying on the TI GUI. 

This method has proven successful with LMK04828, after only a few answers from the kind folks on ti.e2e.com. The DAC chip appears even easier to control, so we are optimistic of getting on top of the chip in short order.

Section 7.3.3 challenges the reader by showing detail of the DACCLK PLL, which is the reference for SerDes PLL, but not the SerDes PLL structure itself. Ultimately, the programmer specifies the SerDes PLL divisors. Absent a diagram of the configuration points (along the lines of LMK manual's figures 12 and 13), it appears that we specify the divisors indirectly, as a function of MPY, Lane Rate (Table 2), and JESD parameter L (perhaps I am overlooking others).

Here is the issue: In Section 7.3.3 "SerDes PLL Output Frequency" appears to mean "F_vco." However, the rows in Table 2 make sense when "SerDes PLL Output Cycle" means "frame clock cycle."

Would you kindly address the inconsistency? It may be that we have overlooked a pertinent detail in the document.

Hopeful thanks --todd

  • Todd,

    Section 7.3.2 and section 7.3.3 are strictly highlighting the SERDES PLL, which is different than the DAC PLL (7.4.12).

    The SERDES PLL will basically take a reference clock (a divided down clock from the DAC PLL or directly from external clock, user selectable from register SERDES_REFERENCE_CLK_SELECT) See below registers

    This reference clock can be divided down by serdes_refclk_div and also multiplied up through MPY value

    The final required SERDES PLL rate depends on your line rate. We typically use full rate. For instance, with 10Gbps of SERDES line rate, your SERDES PLL need to be 2.5GHz.

    -Kang

  • Hi Kang,
    Thank you for this reply.
    Unfortunately, my question pertains only to SerDes PLL, and this reply does not resolve the issue.

    For reference, this design bypasses the DACCLK PLL and divides DACLK by 1 into the SerDes PLL reference.
    Interpolation = 1.

    Follow up questions:
    Q1: Why is the DAC PLL drawn twice, in 7.3.3 and in 7.4.1.2, while the SerDes PLL is not shown at all?
    Q2: Fractional-N PLLs are controlled by divisors. I assume the SerDes PLL is a conventional PLL. Please confirm that this chip infers the divisor values from MPY and Line Rate as above, with no other factors.
    Q3: A naive contextual question: CDR recovers the Lane frequency for the SerDes receiver by analysing the incoming signal. Why does the SerDes receiver require a PLL?

    Hopeful thanks --todd
  • Hi Kang,

    One more follow up question, please:
    Q4: In Table 2, does “SerDes PLL Output clock cycle” in Table 2 really mean “Frame" or, equivalently, "Frame Clock Cycle" or "DACCLK cycle" ?

    Thank you for your kind attention to our issues --todd
  • Hi Todd,

    I understand your concern about the SERDES PLL. Our data converter team obtained the SERDES IP as a whole and integrated the IP with our data converter. Therefore, regarding your question #3, why does the SERDES receiver requires a separate PLL, it is because the entire SERDES is build on top of a single IP.

    Our internal document of the SERDES IP only show 4 major components: power supply regulator for the SERDES PLL, phase frequency detector for the SERDES PLL, VCO circuit, and the MPY. There are no other detailed drawings besides the actual silicon design (which I have no access), and hence indicates a conventional PLL.

    Regarding your questions:
    1. The SERDES PLL consist of very simple elements and hence no diagram was shown. Only the traditional components were shown in our internal documents.
    2. You are correct. The MPY automatically multiplies the PFD directly to the SERDES PLL output rate. This is done automatically. To get the PFD frequency for the SERDES PLL, you will need to divide the DAC PLL or external clock accordingly with the SERDES reference clock divider.
    3. Actually, CDR only changes the sampling phase of the clock to ensure the serialized data stream are sampled at the optimal setup/hold time. The actual sampling clock still need to be generated by a clock source, and this is where the SERDES PLL is needed. The CDR is only a steering wheel, and the actual power still need to come from the engine (i.e. SERDES PLL).

    -Kang
  • Hi Todd,

    The JESD204B standard basically separates the physical layer away from the link layer and transport layer. The physical layer is the SERDES itself (i.e. the mouth and ears of communication), while the link and transport layer are responsible for the handshaking, link establishment, and data packing (i.e. the smarts or the brain of the communication).

    The frame clock cycle or the multi-frame clock cycle are JESD204B standards to establish a common clock cycle in the link layer and transport layer so the data packets are basically processed at the same rate for both the FPGA and data converter. This sort of making the design common so the standards of different vendor can be compliant with each other.

    the SERDES PLL output is strictly used to sample the serialized bit stream. It in theory should have different rate then the frame/multi-frame clock due to the 8b/10b encoding nature of the JESD204B standard.

    The DACCLK cycle is simply the DAC sampling clock for the output core. This is effectively how fast the DAC can generate consecutive samples.

    the JESD204B standard can be available to download on the JEDEC website. We also have various JESD204B trainings available on TI.com. You can do a quick search on these training materials.

    thanks.

    -Kang
  • Dear Kang,

    I deeply appreciate the care you have invested in answering all of the questions that I have asked on the forum in recent days. That level of support is especially valuable for us TI LMK/converters neophytes.

    Fortunately, our team has a deep understanding of JESD204B Subclass-1. My questions pertain to ensuring correct bare-metal implementation of a particular JESD system across a collection of TI parts.

    Your replies to this and the other questions have provided welcome assurance that our team has configured the DAC38J84 JESD RX correctly. I look forward to asking further questions here during the bring-up effort. Thanks again!

    Best regards --todd