subject: A Discrete Pll Requires Analog Design Expertise [print this page] Synchronous clocking is most often used in applications that require continuous communication. Network latency and variability in latency must be minimized. To accomplish this, applications including SONET/SDH, Synchronous Ethernet (SyncE), wireless backhaul, and video transport require that the source and destination operate at the same frequency. On the transmit side, the clocks that provide timing for the transmit path of the high speed SerDes are locked to a highly accurate reference clock. Both primary and secondary reference clocks are supplied from a centralized timing source (e.g. GPS). A PLL is used to lock to this backplane reference, attenuate jitter on the clock signal to remove unwanted noise, and provide a low jitter output clock to the PHY. On the receive side, a Clock and Data Recovery (CDR) unit is used to recover the system clock. This CDR may be an external component or integrated within the PHY depending on the application. A XO may be used to center the CDR to provide fast acquisition and lock. The recovered clock passes through another PLL to divide the clock frequency down to a lower rate. Local timing can either be synchronized to this clock or to another local clock that is synchronized to a centralized timing source. Synchronization to the source ensures clock synchronization across all nodes in the network. Depending on system requirements, the PLLs used in this application may require a low loop bandwidth to filter unwanted jitter from the clock signal.
There are multiple disadvantages with a discrete PLL solution. A discrete PLL requires analog design expertise and is sensitive to board-level noise, so special care must be taken in the design and PCB layout. In addition, a discrete PLL typically provides a single output frequency. If the designs frequency requirements change, a separate VCXO must be sourced. In some applications, multiple VCXOs are required to generate all of the required frequencies in the application, increasing BOM complexity. To address these shortcomings, dual, quad, and even any-rate I2C-programmable VCXOs like Silicon Labs Si571 are available now to address multi-rate applications by replacing multiple discrete VCXOs with a single device.
An alternate approach is to use a jitter attenuating clock multiplier IC, which integrates PLL circuitry on-chip. The clock multiplier maintains lock to the reference clock, filters unwanted jitter and generates a multiplied frequency output clock for the transmitter. Special care must be taken in clock multiplier IC selection, as all clock multipliers are not created equal. For high speed serial data transmission applications, only the highest performance clock multiplier ICs provide the jitter performance necessary to meet the end application requirements. The key specification is maximum jitter generation as opposed to typical jitter generation. A clock multiplier that specifies maximum jitter best enables the hardware designer to allocate a jitter budget among the data path and timing components in the board and ensure there is sufficient margin under all conditions. Also, check the loop bandwidth options available on the clock multiplier IC. If jitter attenuation of the reference clock is required, then oftentimes the loop bandwidth must be 1 kHz or less. Lastly, confirm the clock multiplier IC supports all required frequency plans. An example of a frequency-agile clock multiplier that provides jitter attenuation and any-rate frequency synthesis is the Silicon Labs Si5319.
Lastly, a clock solution is preferable to a discrete solution when system-level clock functions are required. An example of this is hitless switching between input clocks, in which the clock monitors the quality of a primary reference clock and switches to a secondary reference upon detection of an alarm condition on the primary clock. Another popular system-level clock requirement is holdover, in which the clock continues to generate a stable output clock in the absence of a valid referee.