Post Go back to editing

Clock Input Rate and Output Latching

Category: Hardware
Product Number: LTC2387-18

Hello,

We are using the LTC2387-18 for 18 bit A2D conversion using 2 lane mode for an imaging sensor application. 
Each pixel of the analog sensor is being sampled synchronously using a 8 MHz clock which results in a readout window of 125 ns for each pixel. (This is the criteria).
Each pixel's (differential) output is fed into the LTC2387-18's (differential) input. 

In line with the 
LTC2387-18's degrees of constraints and freedom as provided in the datasheet and timing diagram as well as to meet the criteria as mentioned above, we have calculated the input clock to the LTC2387-18 to be 200 MHz as appropriate. (200 MHz seems ideal and optimum but can be managed at 50 MHz and 100 MHz with some difficulty & complications on the FPGA side). The LTC2387-18 shall be operated in 2 lane mode for 18 bit output.   

Based on the 200 MHz input clock to the LTC2387-18 we are sending a CNV signal at the rising edge of the 200 MHz for a duration of 1 clock cycle (5ns) (we can afford to send the CNV pulse for a duration of 25ns i.e 5 clock cycles).

The  following are the queries: 

a) Can the LTC2387-18 handle the 200 MHz clock as input clock
b) Do we explicitly "latch" the FPGA's input onto the DCO output of the LTC2387-18 and would the DCO return the output to LOW (0) after the burst of 5 clock pulses are sent to the ADC?
c) Do we need to read the LTC2387-18's outputs at DA+/- and DB+/- using the latched DCO as the input reference clock?
d) Can the FPGA latch onto the DA+/- and DB+/- of the ADC for the output? (Since the timing diagram shows that the the DA+/- and DB+/- are forced to LOGIC 0 after the output is completed and thereby we presume  that the enforced LOGIC 0 is the default state to bailout from the latched condition) 
e) The D[17] and D[16] bits become available immediately after the mandatory waiting period (allocated for S&H/Buffer/Multiplex etc). Can we read the D[17] and D[16] bits before we send the burst of 5 CLK pulses i.e. before the first pulse of the DCO becomes available?   

Looking forward to the responses for the above> 

Regards

Kingshuk

  • Hi,

    We'll look into this and get back to you.

    Regards,
    Jo

  • Hello Nathan / Jo,

    Eagerly looking forward to your response...

  • Hi  

    At this moment, these are the answers I'm able to provide.

    First, just to clarify, do you intend to use the 200 MHz for the entire system?

    A CNV pulse has a duration of 25 ns, i.e., 5 clock cycles; for this one, there is no problem since the typical tconv is 58 ns and tCNVH is 5 ns. For you to reach the tFIRSTCLK, you need 13 clock cycles, considering the 200 MHz clock input. If you’ll use two lanes, assuming you’ll use five pulses for each to clock out the 18 bits.

    The pulse width of CNV+ should meet the tCNVH and tCNVL specifications in the timing table. It is accurate as long as the time between conversions meets the tCYC specification. 

    It has an internal clock that is trimmed to achieve a maximum conversion time of 63 ns. With a typical acquisition time of 27.7 ns, throughput performance of 15 Msps is guaranteed.

    b. The LTC2387-18 has a serial LVDS digital interface that is easy to connect to an FPGA. Three LVDS pairs are required: CLK±, DCO±, and DA±. A fourth LVDS pair, DB±, is optional, see Figure 11.

    c.The data on DA± is updated by every edge of CLK±. An echoed version of CLK± is output on DCO±. The edges of DA± and DCO± are aligned, so DCO± can be used to latch DA± in the FPGA.

    d. A fourth LVDS pair, DB±, is optional. but if a lower data rate is desired, the two-lane output mode can be used. When the TWOLANES input pin is tied high, the optional LVDS output DB± is enabled, and data is output two bits at a time on DA± and DB±. In two-lane mode, five clock pulses are required for CLK± (see Timing Diagrams)

    e. A conversion is started by the rising edge of CNV+. When the conversion is complete, the most-significant data bit is output on DA±. Data is then ready to be shifted out by applying a burst of nine clock pulses to the CLK± input. The data on DA± is updated by every edge of CLK±. Refer to Figure 12.

    If you have any further questions, feel free to reach out.

    Warm Regards,

    Red

  • Hello  

    Our implementation is as per the attached diagram. We are using 200 MHz only for the ADC's CLK input and referencing the CNV duration and not for the entire system. 

  • Hi  , I think the attached timing diagram is okay.

    Feel free to ask as you proceed with your design.

    Thank you.

    Kind regards,

    Red