Copper-based connectivity has served us well for a long time and will continue to do so in applications where it is effective from a performance and cost perspective. For very-high speed and/or long-distance signaling, however, the material cost and physical signal limitations of using metallic conductors has driven eyes to other transport mechanisms.
Fiber optics is not new, and the telecom industry has pushed development and deployment of fiber-optic transceivers and links so that they now span the globe. Very few of our designs have had the need to traverse long distances at such high speeds. Even fewer of us have had deep enough pockets to set up vast high-speed networks. On the other hand, engineers now are finding that local requirements are pushing the limits of metallic interconnects.
That is because the need for communications is not just long-distant cabling between computer centers. Communication necessities also extend to rack-to-rack and even board-to-board within a box, all of which is pushing fiber-optic technology into more high-speed designs.
Legacy COM links on MCUs were adequate for sub Mbit/sec data rates. Then Ethernet took us from 10 to 100 Mbit/s and now 1 Gbit/s links are commonplace. This is where copper runs out of steam and fiber starts to shine.
This article looks at issues and concerns engineers face when interfacing microcontrollers and fiber optics. This includes the rudimentary tasks of setting up and controlling laser emitter power levels and sensitivity thresholds for receivers, as well as tracking performance in real time. All parts, datasheets, tutorials, and development kits referenced in this article can be found on the Hotenda website.
A view from the top
We can classify the microcontroller marriage to fiber optics into two main categories. On one side are high-end network processors. These implement and decode protocols from multiple data streams at very high speeds, usually in conjunction with an FPGA, and these parts are a class unto themselves.
On the other end are the smaller, dedicated processors that either control or live within (or in close proximity to) modular or discrete transmitters, receivers, or transceivers. Discrete detectors, emitters, lasers, fiber connectors, and cable assemblies certainly let us place all fiber-optic elements on our own boards. The entire communications link can be replicated on a circuit board when only a few links are needed.
Our main embedded processors can certainly store and retrieve calibration data for transmit power levels, receive bias, and thermal compensation. These can be set up sequentially during power-up and reset as part of the initialization process. Tracking performance and dynamically adjusting communications parameters for many links in real-time, however, can take its toll on a central host processor, especially if many fiber-optic links are needed.
A viable approach in this latter case is to design our own module, or modular circuit block that can be implemented and inventoried like any other part and replicated like a copy and paste function on new designs. In this case, small dedicated processors in each module can track and monitor in real-time the links’ performance (Figure 1).
Figure 1: If the central processor has the ability to monitor and correct all the fiber-optic links fast enough, the levels can be maintained at an optimum point. If too few sample and correction cycles take place, parameters can drift to levels that introduce error. One link is not a burden to monitor, unlike 1,000.
At first glance, this may seem like a wasted effort or an unnecessary redundancy. Why have a dozen small processors when I can use just one? This will depend on how many links we need. One link might not take too much host processing time and resources from an embedded controller, but 1,000 links can add up. Small, cheap, and in high volume, preprogrammed dedicated processors in packaged or even die form will be cheaper than most of the other components in your modular fiber-optic design.
Another benefit from this approach is the ability to feed your modular transceivers with an external high-accuracy clock. This allows skew control and synchronized transfers of data across multiple links.
Inside our modules, basic processors can store parameters as well as perform diagnostics. These dedicated processors can also act as serial to parallel and parallel to serial converters when high speed is not a requirement for fiber-optic links.
One key benefit of fiber optics is galvanic electrical isolation. If the box you are making connects to a high-voltage utility line and needs to communicate, Industrial Fiber Optics’ inexpensive plastic fiber optics are an ideal solution, especially if sub-Mbit/s data rates are all that is needed.
Lower-cost plastic fiber optics (Figure 2) are designed for use with industry-common 1.0 mm or 1,000 µm core plastic optical fiber and often use visible light LEDs instead of expensive lasers. Off-the-shelf PCB transmitting diodes as well as receiver phototransistors are easily placed on the periphery of your circuit board. However, most of the time, high data rates will drive the need for fiber, and plastic just will not do.
Figure 2: Low-cost off-the-shelf plastic fiber links are ideally suited for lower-speed microcontroller data rates that can benefit from the high-voltage isolation that plastic fiber provides. These can also run in high-noise environments.
OEM or roll your own?
If our designs need to conform to industry-standard protocols and interoperate with equipment from dozens of different manufacturers, then OEMing a finished and certified module is the fastest and cheapest way to get to market. If you are in control of your link-to-link characteristics, then designing your own will save money if time permits.
The simplest link to control is the discrete transmitter. Transmit-only links are simpler than transceivers and are basically programmable laser drivers switched on and off by data. Discrete laser drivers like the Analog Devices ADN2830ACPZ32
use closed-loop control of average optical power and adjust laser bias to maintain and monitor photodiode current. Bias control ranges from 4 to 200 mA. These can report status, but are typically set via bias resistors or trimming.
Discrete fiber-optic receivers fill out the other end of our transceiver links; and here too, digital-trimpot-style threshold detection and gain selection can allow our dedicated or centralized host processor to control and adjust performance.
Several simple dedicated microcontrollers can be used for our transmit, receive, or transceiver modules, or banks of modules. Basic processors like the Texas Instruments MSP430FR5739IRHAR
are ideal for housekeeping functions inside an optical transceiver module, such as for setting up calibrated power levels from its NovRAM storage. Note the use of FRAM for enhanced reliability. These MCUs can also monitor temperature and compensate for power levels and sensitivity thresholds. TI is actively targeting fiber-optic networking for its processors.¹
Reducing power is a key concern when multiple banks of fiber-optic transceivers are used and TI provides a Training Module on low-power operations and features of the MSP430 family. An overview PTM of the family is also online, as is a focus on design tools. The MSP-EXP430FR5739
development kit that can be used for optical networking is also readily available from Hotenda.
Maxim is another manufacturer who is specifically targeting fiber-optic communications with an embedded controller designed for use with SFP+, QSFP, 40 G, and 100 G optical transceivers, as well as PON diplexers and triplexers (Figure 3). Called an Optical Microcontroller, the DS4830
is a 16-bit microcontroller surrounded by a 13-bit A/D converter with 26 muxable inputs.
Figure 3: The 16-bit embedded microcontroller is dedicated to maintaining the integrity of the high-speed fiber-optic links while optimizing power consumption and controlling active cooling techniques.
A particularly nice feature is the 8-channel, 12-bit buffered DACs along with a 10-channel boost/buck PWM-based DC/DC controller with up to 12-bit resolution. This allows precise and fast power control of multiple lasers rather easily. Settings, adjustments, and overrides can take place via the chip’s I²C port and custom code can be developed for the internal processor through the use of a JTAG port.
Maxim also offers the DS4830
EVKIT# Optical Microcontroller evaluation kit, which supports the entire MAXQ
series of embedded microcontrollers.
As referenced earlier, active temperature and power control are crucial to a fiber-optic link’s reliability. Maxim provides an app note that illustrates the control loop an embedded processor needs to perform to reliably keep temperatures in the safe ranges.²
Fiber is not just for long-haul signaling any more. Modern-day servers need high speed, rack-to-rack communications, and even within a box, signal speeds may dictate the need for fiber-optic communications links.
Whether putting a few links on your PCB, or designing your own modular solution, engineers can take advantage of embedded processors to keep drive and power parameters constant for higher reliability and consistent performance. These processors can also be part of your modular design to save cost or satisfy special needs like high-voltage isolation, or to traverse high EMI/RFI paths without signal degradation.
For more information on parts discussed in this article, use the links provided to access product information pages on the Hotenda website.
- TI: Microcontrollers in Optical Networking.
- Maxim Thermoelectric Solutions