Show me the data High-speed commercial serial buses square off for real-time, military and aerospace applications
Though Mil-Std-1553 has had a long and productive life in military and aerospace electronic systems, the 1-megabit-per-second work-horse no longer offers anywhere near the throughput necessary for today`s information-intensive battlespace.
Show me the data
High-speed commercial serial buses square off for real-time, military and aerospace applications
By John Haystead
Though Mil-Std-1553 has had a long and productive life in military and aerospace electronic systems, the 1-megabit-per-second work-horse no longer offers anywhere near the throughput necessary for today`s information-intensive battlespace.
The demands of real-time sensor-fusion applications, together with order-of-magnitude increases in processing capability and speed have established an entirely new set of bandwidth and speed criteria for the interconnection of subsystems and processors. This combined with the reality of today`s military and aerospace budgets, means real-time system developers are looking to commercially-developed serial data bus topologies for their current and next-generation systems. As a result, the options available to military and aerospace system designers are intimately related to the direction of this broad commercial marketplace.
While there is a common push to close the gap between high-speed multiprocessor systems and local-area networks, divergent market applications are leading to a several different approaches to data buses. "While some data bus network users are primarily concerned with performance and latency issues, others are willing to trade these off against more traditional network concerns such as scaleability and distance," says Craig Lund, and independent industry consultant based in Durham, N.H. Likewise, the appeal of data security and fault- tolerant features also depend on individual applications.
So, like their commercial market counterparts, military and aerospace designers must deal with the challenge of choosing one solution from a host of possible offerings, and almost certainly trading off some level of performance and functionality to take advantage of market-volume cost savings and ensure reasonably long life-cycle support.
Fibre Channel is getting a lot of attention as a high-speed data bus interconnect. In commercial applications, the Fibre Channel Arbitrated Loop (FC-AL) standard is already being implemented in distributed data storage applications while switched Fibre Channel is gaining momentum in expanding connectivity to larger data bus and LAN implementations.
Currently most military and aerospace Fibre Channel users are implementing Fibre Channel chip sets for simple, fast, low-latency, point-to-point communication links. Often, however, these links are application-specific implementations, and beyond this, Fibre Channel has frequently been viewed as a traditional network with too much inherent latency for real-time data bus applications. Now, however an industry working group composed of major defense system prime contractors is specifically focusing on developing Fibre Channel for real-time avionics applications.
The Fiber Channel Avionics Environment (FC-AE) group is working to augment Fibre Channel for avionics by defining a set of real-time constructs for the bus. the FC-AE document, however, will not be a separate Fibre Channel standard, but an operational profile specific to avionics applications, says Michael Hoard, senior project engineer at The Boeing Co. in Seattle. "Fibre channel is already extremely suitable for real-time avionics applications, but some subtle constructs needed to be added," says Hoard, who is also the FC-AE chairman. "The objective of the FC-AE is to streamline the large overall set of Fibre Channel protocols and create a single document specific to avionics implementations."
The FC-AE Fibre Channel physical and signaling interface standards extension (FC-PH3) document is now in the final stages of becoming an American National Standards Institute (ANSI) standard. The document covers three areas: real-time arbitrated loop extension, real-time switching fabric for priority preemption, and real-time isochronous loop.
The real-time arbitrated loop extension, which is intended to replace 1553 with Fibre Channel`s 1-gigabit-per-second shared-media arbitrated loop protocol, incorporates bounded latency protocols and packet-size definitions for getting on and off the loop.
The document also defines a real-time Fibre Channel switching fabric incorporating class 1 priority-preemption protocols and class 6 reliable multitask definitions.
The third area being addressed is a real-time isochronous loop, guaranteeing time-of-arrival in nanosecond increments for precise timing and control applications. Excess bandwidth not used by the periodic isochronous traffic will be available to asynchronous traffic. The protocol allows both bit- and byte level data transmission.
Fibre Channel has already had a measure of success in military and aerospace avionics programs. Recently Ancor Communications Inc. in Minneapolis won a $500,000 Boeing contract for its "GigWorks" Fibre Channel switches for use in the E-3 Airborne Warning and Control System upgrade program while high-speed Fibre Channel PCI Mezzanine Cards from Systran Corp. in Dayton, Ohio, were chosen to switch links.
"Fibre Channel is particularly well suited to the military and aerospace requirement," says Tom Bohman, Systran Fibre Channel product marketing specialist. "First and foremost we`re talking about high-bandwidth data communication for what can be generically viewed as image/signal-processing systems, and Fibre Channel excels at connecting sensors to DSPs [digital signal processors] and DSPs to general-purpose computers and mass storage."
Systran engineers have built a switched variant of the FC-AL standard. Unlike the dynamic inband switching of a switched Fibre Channel fabric, the network-transparent, full-crossbar-matrix switch is for static switching of network configurations to enable designers to implement various combinations of point-to-point and arbitrated-loop configurations. Arbitrated loop is attractive to radar and sonar applications, because "we`re only talking about 3-4 nodes and it is less expensive and easier to implement than a switched fabric," Bohman says.
Military Fibre Channel
Like other commercial architectures, there are no military-qualified Fibre Channel chips, and no particular reason to expect there ever will be. "When more than 50 percent of military applications can get by with upscreened commercial products, and the full-mil market shrinking, we can`t compete on price," explains Anthony Jordan, Standard Product Line Manager at UTMC Microelectronic Systems in Colorado Springs, Colo.
Experts at DY 4 Systems Inc. of Kanata, Ontario, however, are evaluating commercial Fibre Channel devices for use in a planned project that involves military-grade electronic components. Company officials are making good progress in identifying the potential problem areas of commercial Fibre Channel chips and are confident they will have a flight-qualified solution before 1999, says Duncan Young, DY 4 director of marketing. "We feel strongly that Fibre Channel will be one of the technologies to replace 1553, particularly in avionics applications which need very high-bandwidth," Young says.
DY 4 officials expect flight-qualified Fibre Channel products will find significant opportunities in future avionics upgrade programs such as the Boeing F/A-18 fighter-bomber. Young points out that engineers can implement Fibre Channel in either copper wire or optical fiber, which enables designers to mix and match switched and loop topologies. Young also says he believes Fibre Channel`s high-level protocol independence will be an attractive feature for developers. DY 4 officials plan to announce a commercial partnership within the next one or two months for their extended-temperature Fibre Channel products.
Asynchronous Transfer Mode
Originally developed for the telecommunications industry, asynchronous transfer mode (ATM) network technology is another candidate for widespread data bus network implementations. ATM is an extremely versatile technology that can easily be made suitable for military and aerospace data bus applications, says Philippe Vincent, vice president of marketing & communications, at Cetia Inc. on Toulon, France.
Because of its widespread telecommunications base, ATM should also offer lower-cost points than other data bus solutions. Currently most ATM implementations run at 155 megabits per second, but the specification also addresses low-cost systems with speeds as slow as 25 megabits per second.
Industry engineers also are working 50 megabit-per-second interface over plastic optical fiber as well as a 622-megabit-per-second variant. Cetia designers have implemented 155 and 25 megabits per second data rates in their "CPMC-ATM" PMC boards for PowerPC central processors at Thomson Marconi Sonar for real-time sensor systems in an Australian Navy shipboard application. Cetia experts wrote the software for the modules including LAN emulation, classical IP, RPC1483 IP, and ATM Forum UNI3.1 signaling and network management utilities. Cetia officials expect to have a 622 megabit-per-second rate product available next year.
Some observers, however, say they believe ATM has lost a lot of its early momentum as a candidate for widespread data bus solutions, and say they see ATM being pushed back into the telecommunications/wide area network world.
While Cetia`s Vincent acknowledges that some of ATM`s potential has not yet been developed, he points out that "ATM standards are now quite well satisfied in terms of hardware," and that what remains is the software development necessary for specific applications and diverse requirements. "Since software development must be addressed on a requirement-specific basis, we need customers to help sponsor this development work." Still Vincent says he recognizes that ATM is relatively new to many data bus network applications, and acknowledges that it will be difficult for ATM proponents to penetrate these markets.
Lund says he agrees that ATM "still has some real play in very large platform applications like ships and large surveillance aircraft platforms that tend to use more wide area network technology." The relatively benign operating environments of these platforms also makes the use of off-the-shelf commercial-grade chips and products more viable than they are for tougher platforms such as aircraft and combat vehicles.
Vincent says Cetia`s approach to demanding military and aerospace applications, however, will be to ruggedize at the board level, often with the company`s conformal VME board shell that Cetia marketers call "the Ruggedizer."
Designers base several real-time military and aerospace systems and products on a technology generically called reflective memory. In general, reflective memory works by duplicating (reflecting) writes to the memory board of one computer system onto a parallel memory board in a second system over a dedicated interconnect line. Engineers use a synchronization primitive to ensure that the two boards remain in sync.
One such system is Systran`s fiber-optic Shared Common RAM Network, better known as SCRAMNet. This interconnect is popular for military simulation applications as well as ground-based telemetry and real-time control environments. Systran designers optimize SCRAMNet for transferring small packets of data at 150-megabit per second. Data latency per node is from 250 to 800 nanoseconds with the SCRAMNet protocol designed to avoid network collisions and software delays.
Most designers do not consider reflective memory, however, as appropriate for data bus-type applications. "Although reflective memory systems are good solutions for relatively small-scale clustering applications (very small number of very large systems), they don`t scale at all well," Lund explains. As systems scale up, complex, time-consuming broadcast protocols are necessary to manage the network. Since duplicate memory boards are required for each node in the network, reflective memory is also expensive to implement in large systems.
SCI - sharing vs. passing
Representatives of another industry group believe the Scaleable Coherent Interface (SCI) is the best approach for military and aerospace applications - particularly for avionics networking. SCI, which became an approved ANSI/Institute of Electrical and Electronics Engineers (IEEE) standard in 1992, is a cache-coherent shared-memory architecture using copper or optical fiber as the physical medium. Though the hardware is more complex than it is for reflected memory systems, cache-coherent, shared-memory systems are much more scaleable and faster architectures.
Typical SCI performance is in the 200 megabyte-per-second range (CMOS), but the SCI specification covers speeds as fast as 1 gigabyte per second (BiCMOS/GaAs) over distances of tens of meters for electrical cables and kilometers for serial optical fiber. Designers at IBM have already demonstrated a single-chip SCI interface running at 1 gigabyte per second.
SCI proponents point out that software programming costs more than hardware in military and aerospace systems today, making SCI`s shared memory approach a more productive paradigm than message passing. Because SCI eliminates the need to translate layers of run-time software protocols, it also reduces programming costs and interprocessor communication delays compared to systems based on networking and I/O channel protocols such as Fibre Channel and ATM.
"SCI`s very low latency (2.7 microseconds in current implementations measured at application level) make it an ideal choice for military and aerospace network architectures," says Gary Roberts, director of marketing for Dolphin Interconnect Systems of Framingham, Mass. Dolphin makes SCI controller chips and plug-in boards supporting PCI and S-bus. Using CMOS technology, Dolphin engineers build products that deliver data rates as fast as 500 megabytes per second.
Systran officials are also using the federal Small Business Innovative Research program to look at SCI as a means to connect PowerPC microprocessors across an SCI backplane. "In general, where you can have a parallel switched system, SCI`s shared memory communication can be very efficient," Bohman says, adding, "When you go point-to-point with pure serial communication, Fibre Channel is farther ahead than SCI."
Ralph Lachenmaier, senior engineer at the Naval Air Warfare Center at China Lake Naval Air Warfare Station, Ridgecrest, Calif., is Chairman of the SCI Real Time (IEEE 1596.6) group working toward a unified real-time SCI network. The group`s goal is to make SCI suitable for backplane and point-to-point LAN applications. "SCI is the first protocol to be able to do this, and unlike other multipoint buses, which sacrifice speed for distance and scale, SCI is relatively distance insensitive," Lachenmaier says.
In addition to working on base SCI, members of the real-time group are teaming up with Firewire-bus developers (see story page 22) on the next-generation "Serial Express" standard which combines Firewire`s isochronous mode with SCI`s point-to-point capabilities for real-time, fault tolerant, high-security applications. "We recognize that Firewire`s isochronous mode has some uses in the military world and our two groups are closely integrated," Lachenmaier says. In fact the SCI real time group has been somewhat stalled "waiting to see what will come out of serial express," he says.
Lachenmaier acknowledges that SCI has not yet had sufficient commercial interest to bring prices down. Yet balancing this trend, he predicts, will be SCI`s support for shared memory and byte-addressable message passing, while Fibre Channel and ATM support only message passing.
SCI`s shared memory approach offers lower latencies than message passing systems, Lachenmaier says. He cites benchmark studies at the University of Florida to highlight SCI`s latencies of about 1 microsecond one-way, while Fibre Channel was in the millisecond range, mostly because of software overhead. "SCI allows you to put equipment where you want it, giving you access and weight distribution with backplane speed," Lachenmaier says.
Not everyone agrees, however, that shared memory is necessarily a better approach than message passing. Pointing to the high power consumption and heat generation of shared memory hardware, even some users of SCI do not implement its shared memory layers. Other parallel computer developers - particularly those working with signal processing - prefer not to deal with the added complexity and high-traffic rates of shared memory hardware.
In response, Lachenmaier points out that designers can implement SCI either in cache-coherent shared memory, in non-coherent shared memory, or in message-passing mode. "In message-passing mode, SCI`s heat generation is no greater concern than other buses, and in non-coherent shared memory mode, you still won`t have a lot of heat generation," he says. "While first generation gallium arsenide devices created a lot of heat, high-speed devices are now available in CMOS."
Joint Strike Fighter
The U.S. Navy/Marine Corps/Air Force Joint Strike Fighter (JSF) has emerged as a milestone program for the future of commercial bus architectures in military and aerospace applications, with most industry and standards groups expecting the program`s eventual choices to guide decisions of other real-time network developers.
Unprecedented demand for high-performance data fusion has made the avionics network of the next-generation JSF a crucial element of the program. "Enhancements in situational awareness will require a paradigm shift in the way in which we manage information in the cockpit and a revolutionary approach to the distribution of intelligence, surveillance, and reconnaissance data," state JSF requirements.
Aircraft designers from Boeing and Lockheed Martin are developing concept demonstration designs for the JSF. Though the program will ultimately build three different JSF designs, the aircraft will have common avionics architectures.
The JSF program has focused several data-fusion activities, among them the JSF On-Board/Off-Board Information Fusion and Management Study, conducted at Lockheed Martin. This included an evaluation of the Scaleable Coherent Interface/ Real-Time (SCI/RT) architecture. The study determined that "efficient on-board/off-board information fusion and management will be critical for the JSF`s single-crew cockpit."
Since the JSF must also address affordability concerns, a parallel Integrated Core Processing technology maturation project focused specifically on "achieving an affordable JSF avionics suite through an open systems architecture leveraging COTS processing capability."
An initial assessment of unified network protocols for JSF avionics has been completed, and "an initial list of candidate protocol standards has been identified by the JSF Open Systems Architecture Joint Government/Industry Integrated Product Team," says Air Force Maj. Dan Bohr of the JSF Program Office in Arlington, Va. Among the standards assessed were ATM, Fibre Channel, FC-AE, Gigabit Ethernet, Myrinet, SCI, SCI-RT, Serial Express, and S-Connect. "The assessments will be detailed in the JSF Architecture Definition (JAAD), Version 2.0, which will be released this fall," Bohr says. "The candidate standards and accompanying assessments will then continue throughout the JSF concept demonstration phase."
No matter which data bus architecture military and aerospace developers select for their real-time applications, they will have at least a few shortcomings to deal with. In addition to whatever system or application-specific challenges they must design around, when it comes to selecting their real-time network or bus topology, there will only be a best solution as opposed to a perfect one, and the scale continues to slide and bend.
Successful implementation of commercial data bus technology "can be done, but the overall problem with commercial bus technology, is the lack of staying power," says Chris DeLong, Honeywell senior staff engineer and chairman of the Avionics Systems Division of the Society of Automotive Engineers (SAE). Observing that the F-22`s linear bus development work began 12 to 15 years ago and the platform is still not in production, DeLong adds "By the time the DOD can get on board with a particular commercial architecture, it`s already obsolete or no longer supported."
Although DeLong acknowledges that the linear bus is not a cheap interface, and that military leaders have no desire to build another mil-std bus, he cautions that "the utility of standard commercial architectures must also be weighed against their ability to meet the avionics requirement. The problem is some of these buses won`t."
Ultimately UTMC`s Jordan says he doesn`t believe any one particular bus or topology will ever be adopted as a standard solution. "Decisions will be made individually on a cost/performance basis and system designers will use whatever best meets their particular needs at the time."
A high-speed linear serial data bus capable of 100 megabits per second is part of the U.S. Army Boeing-Sikorsky RAH-66 Comanche scout-attack helicopter, which must process and display massive amounts of data through its cockpit avionics, pictured above. The Comanche`s data bus will be able to move data over optical fiber or copper cable.
Fibre Channel is fast emerging as the high-speed data bus of choice for U.S. military avionics. The U.S. Air Force Boeing E-3 Sentry AWACS aircraft, pictured above, is set to receive the Fibre Channel high-speed bus in an upcoming upgrade program.
The U.S. Navy/Marine Corps/Air Force Joint Strike Fighter, depicted above, is the centerpiece for future of commercial bus architectures in military and aerospace applications. An initial list of candidate protocol standards for the JSF include ATM, Fibre Channel, FC-AE, Gigabit Ethernet, Myrinet, SCI, SCI-RT, Serial Express, and S-Connect.
Distributed vs. centralized processing
A basic question in selecting future avionics buses: must individual processing elements be housed in one central location or distributed throughout the airframe?
"There`s no longer a need to connect a lot of widely spaced black boxes in either military or commercial avionics environments," explains Chris DeLong, Honeywell senior staff engineer and chairman of the SAE`s Avionics Systems Division.
DeLong points out that central racks of plug-in modules are replacing the distributed line-replaceable units of conventional avionics systems in the U.S. Air Force F-22 Raptor advanced tactical fighter and Army RAH-66 Comanche scout/attack helicopter, as well as in the new Boeing 777 civil jetliner. In the F-22, for example, all the integrated modular avionics are in two racks connected via a Pi-bus interface.
High-speed serial data moves over the F-22`s variable-speed, fiber-optic linear bus which can run as fast as 100 megabits per second on coaxial copper cable or optical fiber. This linear bus is also part of the RAH-66, and "will meet the high-speed data handling requirements of both aircraft for the foreseeable future," DeLong says.
Manufacturing the military-qualified, high-speed data bus components for both aircraft are engineers at the Government Aerospace Systems Division of the Harris Corp. Electronic Systems Sector in Melbourne, Fla.
While there are advantages to centralized processing, there are also drawbacks, experts point out. "With centralized processing, all your sensor and actuator inputs have to come to one central point and that means big bundles of wire," says Tom Bohman Fibre Channel product marketing specialist for Systran Corp. in Dayton, Ohio.
"Fibre Channel is particularly well suited to the military and aerospace requirement," Bohman says. "Also for many functions, it just makes more sense do most of the A-D conversion and processing close to the antenna, passing only the required information via the data network."
Ralph Lachenmaier, senior engineer at the Naval Air Warfare Center at China Lake Naval Air Warfare Station, Ridgecrest, Calif., observes that although it may have once been necessary to pack processing elements closely together to attain fast data throughput, this is no longer the case with today`s new point-to-point buses and networks, such as SCI. "Few systems have less than 50 nanoseconds bus latency requirements," Lachenmaier says. "That means the jury is still out as to what the best approach will be."
While acknowledging that new platforms such as the F-22 and Joint Strike Fighter may have different architectural models, Duncan Young, director of marketing at DY 4 Systems Inc. of Kanata, Ontario, says he doesn`t see the avionics community moving totally away from distributed architectures. "One of the biggest markets over the next few years will be in upgrades to existing airframe avionics like the F-18," he says. "These will use a mixture of federated and integrated architectures." - J.H.
Emerging buses and standards
In addition to the family of data bus architectures already implemented in real-time, or near-real-time military and aerospace applications, several other candidates soon are expected to join the fray. Two of the most often discussed approaches are Gigabit Ethernet and IEEE 1394, better known as Firewire.
Gigabit Ethernet (IEEE 802.3z), approved in 1995, is a 1-gigabit-per-second extension to the 10-megabit-per-second and 100-megabits-per-second IEEE 802.3 Ethernet standards.
Sponsored by the Gigabit Ethernet Alliance in Cupertino, Calif., the extension will support new full-duplex operating modes for switch-to-switch and switch-to-end-station connections and half-duplex operating modes for shared connections using repeaters and the CSMA/CD access method. Initially operating over optical fiber, Gigabit Ethernet will also be able to use Category 5 unshielded twisted-pair cabling and coax.
Current efforts in IEEE 802.3z draw heavily on Fibre Channel and other high-speed networking components with initial implementations of the standard employing Fibre Channel 780-nanometer (short wavelength) optical components for signaling over optical fiber and 8B/10B encoding/decoding schemes for serialization and deserialization.
IEEE 1394 (Firewire) is a scaleable, "hot-pluggable," serial interface standard for transporting data at 100, 200, or 400 megabits per second and is now also working toward a 1-gigabit-per-second data rate. Sponsored by the 1394 Trade Association (c/o Adaptec, Austin, Texas), Firewire was originated by Apple Computer as a desktop LAN and subsequently developed by the IEEE 1394 working group.
A flexible topology, Firewire supports daisy chaining and branching for true peer-to-peer communication with bus management built upon the IEEE 1212 standard register architecture. For military and aerospace system developers, an important feature of Firewire is its support of asynchronous and isochronous data transfer. The isochronous data channels provide guaranteed data transport at a pre-determined rate with just-in-time delivery to eliminate the need for costly buffering.
IEEE 1394 has been accepted as the standard digital interface by the Digital VCR Conference and has been proposed as an international standard. Members of the EIA 4.1 subcommittee have also voted for IEEE 1394 as the point-to-point interface for digital TV as well as the multi-point interface for entertainment systems.
The European Digital Video Broadcasters have endorsed IEEE 1394 as their digital television interface as well, and several member companies have proposed IEEE 1394 to the VESA (Video Experts Standards Association) for digital home network media.
New extensions to 1394 include gigabit cabling speeds, 100-megabyte-per-second backplane implementations, longer distances using copper and optical fiber, as well as gateways to other communication interfaces such as ATM.
1553 data bus lives on
Mil-Std 1553 was originally released in 1973 and adopted by the DOD in 1978 as Mil-Std 1553B. Nearly 25 years later 1553 is clearly a long-in-the-tooth relative to commercial bus technologies, yet the standard is still far from the scrap heap.
"There hasn`t been a large groundswell of demand for greater bus bandwidth in standard avionics flight, weapons, and environmental control," says Anthony Jordan, standard product line manager at UTMC Microelectronic Systems in Colorado Springs, Colo.
"1553 will continue to be lucrative for at least another 5 years, driven by retrofit and precision munition applications," Jordan says. In fact, UTMC officials have just introduced a new 1553 interface product for low-cost avionics applications such as precision munitions, which they expect to begin producing by this spring. The "Summit" 1553 remote terminal interface integrates transceivers and 4K-by-16 memory in a 1-by-1-inch package.
While SAE`s DeLong agrees that commercial point-to-point buses like Firewire may already be good replacement candidates to connect specific sensor systems, "for the majority of boxes, (communications, navigation, stores management, etc.), 1553 will still be more than adequate," he says.
Further extending the useful life of 1553 may be modest upgrade programs now in the works. SAE Working Groups are developing 1553 extensions for 3-to-5-megabit-per-second data rates using existing cabling, and as fast as 20 megabits per second when combined with upgrades to copper or optical fiber.
Who`s who among high-speed data bus providers
Ancor Communications Inc.
6130 Blue Circle Drive
Minnetonka, Minn., 55343
150 rue Marcelin Berthelot
ZI de Toulon-Est BP244
83078 Toulon Cedex 9
Phone: +33 4 94 16 34 00
Fax: +33 4 94 16 34 01
Dolphin Interconnect Solutions
959 Concord St.
Framingham, Mass., 01701-4682
Phone: 508 875-3030
Fax: 508 875-1517
DY 4 Systems Inc.
333 Palladium Dr. M/S 252
Kanata, Ontario, Canada K2V 1A6
Minneapolis, Minn., 55440
4126 Linden Ave.
Dayton, Ohio, 45432-3068
Sales Phone: 800-252-5601
UTMC Microelectronic Systems
4350 Centennial Blvd.
Colorado Springs, Colo., 80907