Evolution and Adoption of Commercial Technologies for Avionics I/O
Most military aircraft have two major avionics systems which today are only tenuously linked to each other – the flight control system and the mission system. The flight control system includes everything that is required to fly the aircraft: cockpit instrumentation, air data, inertial systems, engine control, electrical power generation, hydraulics, fuel systems, autopilot, navigation, GPS, ILS, landing gear, flying surfaces, slats, flaps and so on. Primary flight control may be augmented or fully digital fly-by-wire, introducing further complexity into the flight control system.
Civil aircraft also have two systems but they are not really linked: the flight control system, and the passenger comfort and entertainment systems. Stringent safety requirements have always lead the avionics industry to believe in the need to develop their own standards for I/O interfaces. This lead to the creation of ARINC standards for many types of data transmission. However, economic reality plus the complexity of new system requirements will make it increasingly necessary to leverage commercial technology in lieu of ARINC or other standards in the future.
Balancing avionics with COTS
Architecturally most current systems are federated, meaning that there are lots of small subsystems linked together to a central computing complex which, in the military example, is either the mission computer or the flight control computer. ARINC 429, ARINC 629 and MIL-STD-1553B are today's industry standards for federated architectures with tens of thousands of implementations in active use. These avionics interfaces are used to instrument all the parameters of the various subsystems and electronically collect this data for presentation to the aircrew in order to fly the aircraft. ARINC 429, ARINC 629 and MIL-STD-1553B provide deterministic, inherently safe, multiply-redundant means of data distribution that can be modeled and verified for type certification
Future systems, both commercial and military will be much more integrated in order to share and process data more effectively and, at the same time, reduce weight and overall cost. However, despite the convergence by both military and civil avionics on commercial technologies for their next generation systems, each has chosen a different technology path and each find it impossible to use their preferred technology without making some industry-specific tweaks. AFDX (ARINC 664) is slated for adoption by a number of future large commercial aircraft, such as the Airbus A380, while military combat aircraft are adopting Fibre Channel. The market will determine whether it is possible to sustain these technologies for the projected life of the airframes.
By looking at each of the standards in turn, it is possible to see how systems are evolving and how commercially-derived data processing architectures and technology can be leveraged. Changes have to take place within the demonstrable constraints of safety, determinism and robustness that are inherent characteristics of avionics applications.
ARINC 429 is one of the longest-lived standards and is in use throughout the civil aircraft market ranging from helicopters, general aviation and commercial passenger aircraft. ARINC 429 makes single point-to-point connections using serial data transmission over a twisted pair. Data is self-clocking at either 12.5 KHz or 100 KHz and one twisted pair is used in each direction. One transmitter can drive up to 20 receivers. ARINC 429 is used to interconnect discrete subsystems in a federated architecture so that, as a simple example, an inertial reference system (transmitter), an air data computer (transmitter) and a radio altimeter (transmitter) would feed their data on a regular basis to an autopilot system (three receivers). In a realistic configuration each of the transmitters would be feeding data to many receiving subsystems and paths would be duplicated or configured in triplicate for safety and redundancy.
An ARINC 429 transmitter has two basic states: either transmitting null clocks or a 32-bit data word. Data words are separated by null clocks. The 32 bits of the data word are separated into fields with, generally, bits 11 through to 29 being the data content. The data content describes the data type, format and parameters of the transmission. This data content and its frequency of transmission is mainly defined by the ARINC standards. It consists of labels (identifying the parameter being transmitted), the source of the transmission, the range (or scale), and the number of data bits used to describe the parameter and the resolution. In addition, the data word can be used for subsystem fault reporting and diagnostics, plus custom functions defined by the many individual manufacturers of avionics subsystems. For embedded systems development and deployment, the SBS Technologies A429-PMC, illustrated in Figure 1, is an example of an easily integrated, mezzanine I/O module providing 8 receive and 8 transmit channels of ARINC 429.
(Click graphic to zoom by 1.6x)
ARINC 575 pre-dates ARINC 429 and is similar electrically and functionally, allowing the same hardware to support either, but 575 is limited to 12.5 KHz clocking and didn't offer the parity bit in the data word. With the development of more intelligent subsystems it has become necessary for ARINC 429 to support the transfer of files between subsystems. The Williamsburg protocol was developed to provide a negotiated file transfer mechanism supporting file sizes ranging from 3 to 255 words in length. Although ARINC 429-based implementations are reliable and intrinsically safe, they require a considerable amount of discrete point-to-point cabling around an aircraft which is both costly and heavy.
Newer replacement technologies, such as ARINC 629 and AFDX, have been developed which significantly reduce this overhead by replacing all the discrete point-to-point cables with the equivalent number of virtual channels on a four-wire network. At the same time, these newer standards mimic the architectural simplicity of an ARINC 429-like federated architecture. This virtual simplicity is essential for when the completed aircraft is submitted for final type certification by the FAA or other civil or military authority.
Developed by Boeing for the 777, ARINC 629 has found few other applications. It is a time division multiplex serial bus using Manchester bi-phase coding, clocked at 2 Mbps. Multiple receivers and transmitters are attached to the same bus using either current or voltage coupling. Like Ethernet, ARINC 629 uses a carrier sense, collision avoidance mechanism to avoid contention when more than one transmitter wants to use the bus at the same time.
When contention occurs each terminal has three timers that are used to back off the next try for variable periods of time. Data is transmitted in messages made up of a number of 20 bit words. Every receiver listens to the ARINC 629 bus and reads the first word to determine if the message contains data that it requires, allowing a simple multi-casting mechanism. Unwanted data is discarded. Each terminal attached to the bus has a unique, fixed personality which defines its function, what data it will transmit, the transmission frequency and what messages it will receive.
AFDX (ARINC 664)
Avionics Full-DupleX switched Ethernet was developed by Airbus SA for the A380, in an attempt to leverage COTS technology for high integrity avionics applications. It is based on 100 Mbps switched Ethernet with some modifications to help combat jitter and a lack of determinism that would be inherent in an unmodified Ethernet implementation. Architecturally AFDX uses a duplicated full-duplex, switched network to reach around the aircraft. Full-duplex is used in order to avoid collisions and retransmissions that are found in a half-duplex Ethernet implementation, thus ensuring that the connection is always available when data is ready for transmission. In this way jitter is reduced under high traffic conditions across the network. However, as the switches use store-and-forward technology there will still be a small but quantifiable amount of jitter due to the receiver operating asynchronously from the transmitter.
Many aircraft subsystems are still very basic, requiring little processor power and generating data at very low rates. It would be uneconomic to hook each subsystem to AFDX directly so they are grouped physically and connected to AFDX by what is known as an “end system”. An end system can be thought of as analogous to a remote data concentrator with a network interface. It has at least two full-duplex (separate send and receive pair) connections to the central AFDX switches. An end system can communicate with any other end system via the switches. Subsystems communicate with each other over the AFDX network using virtual links. The configuration of these virtual links is fixed and analogous to an ARINC 429 implementation having a data source (such as inertial reference system, air data computer, radio altimeter or GPS from the earlier ARINC 429 example) feeding to one or any number of receivers.
Because AFDX uses only these virtual links, the Internet address field of the Ethernet packet is unused, being replaced by a virtual link field. A virtual link defines the data source and its destinations. To further improve jitter and determinism of the final system, each of the virtual links can have a defined bandwidth set by its transmission frequency and its packet's maximum payload size. By fixing these parameters, the data flows and behavior of the network can be modeled accurately during design and can be verified when the system is validated for type approval.
Originally developed by the military for use in tactical aircraft as an instrumentation bus, MIL-STD-1553B has been adopted in many hundreds of projects including ground vehicles and naval vessels with diverse uses. Many of the installations are very different from its original design intentions. MIL-STD-1553B is a time division multiplex bus that uses Manchester encoding, clocked at 1 MHz. Implementations use dual-redundant buses for integrity and protection from battle damage.
Each 1553B bus has a bus controller and up to 32 remote terminals which can be further divided into 32 sub-addresses. A remote terminal or a sub-address may be a subsystem with a specific set of functions such as radio altimeter, GPS receiver or missile warning receiver. Bus operation is strictly controlled by the bus controller – there is no contention for the bus as all receive or transmit operations by the terminals are initiated by messages sent by the bus controller. This is a very robust, deterministic system that is ideally suited to periodic polling of many self-contained, intelligent subsystems. There are many device-level implementations of 1553 which incorporate all the functionality needed to implement a bus controller or remote terminal in a very small physical space. Figure 2 illustrates a 3U CompactPCI MIL-STD-1553B interface from SBS Technologies called the 1553-CPC3 that uses state-of-the-art FPGA technology to implement three complete dual-redundant MIL-STD-1553B channels in a single, 3U-size card slot.
MIL-STD-1553B, like ARINC 429, resulted from the development of federated avionics system. However, as its application base rapidly expanded the standard’s inherent robustness and made it a leading contender for use in mission systems. As a result, this encouraged the integration of flight control with mission systems in military applications. The relatively low bandwidth of the bus prevents sharing raw data, imposing an architectural style on mission systems built around 1553. This manifests itself as many semi-autonomous, intelligent subsystems disseminating highly processed data such as radar target tracks or missile warnings to a central computer.
This central computer, in turn, assists the pilot or crew in their tactical decision making. As sensors have become more effective and systems have developed the ability to engage multiple targets simultaneously, the limitations of 1553 in this role are being exposed. Where much raw data is to be shared between subsystems it has become necessary to supplement 1553 with dedicated high speed links for sensors such as radar, infra-red and TV. However, faster versions of 1553, including optical, have been developed which may improve the situation for legacy system upgrades - but will not reach the GHz+ rates being demanded of new systems. [Editor’s note: look to the August 2005 issue of VMEbus Systems for an article forecasting the extension of MIL-STD-1553 to speeds exceeding 100 Mbps.]
Like AFDX, Fibre Channel represents another successful attempt to leverage commercial technology for military use. It was developed as a commercial network storage technology providing very fast data rates between storage devices, mainframes and workstations. It can simultaneously support multiple protocols such as SCSI and IP to provide great flexibility when hooking together many different types of devices. It is being used as a replacement for MIL-STD-1553B in advanced platforms such as the FA-18E/F Super Hornet and the F-35 Joint Strike Fighter. Fibre Channel operates at 2 GHz (optical) or 1 GHz (copper) and supports three major topologies: point-to-point links, arbitrated loop or, in conjunction with a switch, as a switched fabric. Ports to Fibre Channel are bi-directional and support the full data rate simultaneously while receiving and transmitting.
Fibre Channel is best suited to the implementation of complex, high performance mission computing systems with data from many sensors to be fused into a tactical picture and to control many different types of weapons systems because of the flexibility and performance potential of its architecture. Fibre Channel can be used without a switch as a Local Area Network (LAN) between a number of processors in the same or adjacent boxes using the arbitrated loop topology. However, for connection to the sensors and weapons systems, a switched fabric provides higher bandwidth and deterministic access to the fabric. It is possible to integrate the flight control systems into a Fibre Channel switched fabric. This would be done in the same way as for AFDX where an economic number of I/O subsystems would be connected to access points on the fabric.
There is a continuing debate about time-triggered versus event-triggered protocols for avionics applications. Time-triggered means that data is sampled at regular periods and transmitted to the host. ARINC 429 is a good example of a time-triggered system. Event-triggered means that data is only sent when something changes. Many multi-tasking real-time operating systems work this way, switching tasks when triggered by interrupts.
CANbus, which is seeing widespread application in the automotive industry, has good bandwidth, can be configured to operate in triple and quadruple redundant systems, is readily available and cheap, and would seem an obvious candidate for adoption in avionics. However it is inherently event-triggered and has not yet been selected for widespread avionics use. A new development is TTP (Time-Triggered Protocol) from TTTech which is slated for wide adoption in avionics and automotive applications as an alternative to CANbus. TTP can also be used in conjunction with AFDX as the connection between a number of small subsystems and an AFDX end system.
AFDX and Fibre Channel represent state-of-the-art today in their respective marketplaces and a considerable investment of resources by their devotees. If the current rate of momentum continues, it may just be possible to leverage only one commercial technology for the future, without avionics-specific modification, by both military and commercial avionics.