The dawn of CAN

25th anniversary

CAN in Automation was founded 25 years ago, but CAN is even older than
that. It’s been a long way from CAN’s development in 1984 to its recent
update to CAN FD.

Figure 1: SAE paper 860391 (Photo: Bosch)

I n the eighties, the functionality of automotive systems

was greatly improved by the introduction of electronics.
Electronic engine management impressively reduced fuel
controllers in vehicles. These systems do not have real-
time capabilities and were too expensive.
To bridge this performance and cost-related gap,
consumption and exhaust emissions at the same time, Bosch developed a new serial communication protocol
while ABS optimized braking distance and improved vehi- specifically with a view to in-vehicle data transfer: CAN.
cle control. Following improvements in automotive elec- The main part of the development was done in 1984 and
tronics required linking different functions together, which was immediately followed by the design of the first CAN
operated separately before that. At that time, the intercon- implementation, in cooperation with Intel. The first publica-
nections needed a plurality of separate signal lines. The tion on CAN was presented in February 1986 at the SAE
complexity of the cabling increased costs and caused dif- conference in Detroit. At the same conference, Bosch and
ficulties for the conventional electrical connections regard- Intel released a press statement
ing space limitations, reliability, and accessibility. By 1990, announcing the first CAN con-
high-end vehicles had up to 100 cable connections to the troller IC. In 1987, the Intel 82526
dashboard. This rising intricacy was barely manageable. A consisted of 30 000 transistors,
fast serial communication link was needed to provide suit- had a size of 20 mm² in a 44-pin
able communication between real-time controllers, sen- package and was produced in
sors and actuators in cars, transmitting the information in a 1,5 µ CHMOS III technology. It
coded form, while requiring as few connector pins as pos- was a so-called full-CAN con-
sible. The conditions in the vehicle make specific require- troller, meaning that it stored
ments of a bus system: received messages in dedicated
Bus access priority must be granted dependent upon messages buffers, depending
the importance of the information to be transferred. on the results of acceptance
Short transfer times, high error immunity, and a com- filtering.
puter load as low as possible must be guaranteed. CAN was an immediate
A large number of different messages must be able to success, several other IC imple-
be processed. mentations followed soon (e.g.
The overall costs of the system should be as low as Philips, Motorola), and it was
possible. integrated into µCs. The second
The serial communication systems that were known CAN IC (Philips 82C200) was a
at that time only partially met the requirements of the auto- so-called basic-CAN controller,
mobile industry. Simple interfaces such as UART (Uni- meaning that it stored received
versal Asynchronous Receiver/Transmitter) did not offer messages in a FIFO. The con-
adequate performance. Expensive and complex communi- formity of all CAN implementa- Figure 2: Typical CAN
cation systems such as those used to couple mainframes tions was helped by standardiza- wiring harness (Photo:
(Local Area Networks) were not suitable for coupling tion (ISO 11898) and by Bosch's Bosch)

8 CAN Newsletter 1/2017

 This was also the reason for the founding of CAN in Auto-
mation 25 years ago: an organization for the joint develop-
ment of standards around CAN and a meeting place for
users and suppliers of CAN silicon, tools, and applications.
CiA first standardized CAL (CAN application layer), which
was the basis for the European research project ASPIC
that developed CANopen. CANopen is now maintained
by CiA and has been established as the main higher-layer
protocol for CAN in industrial applications, leaving only
minor roles for its predecessors.
The rapid spreading of CAN into other applications,
combined with the long development cycles in the auto-
Figure 3: Joint motive industry, had the astounding consequence that
press release (Photo: Bosch) the number of CAN nodes produced in industrial automa-
tion was larger than the number of CAN nodes in cars,
simulative reference model for verification. The first CAN until CAN was introduced into the high-volume cars in
evaluation boards and design tools arrived with the first sil- the mid-nineties. The experience from the first CAN net-
icon, allowing everyone to experiment with CAN networks. works merged into two updates of the CAN protocol: CAN
The performance, robustness, and simplicity of CAN, 1.2 (1990) increased the oscillator tolerance, allowing the
combined with the multi-sourced availability of CAN con- use of ceramic resonators and CAN 2.0 (1991) introduced
trollers, created interest not only from the originally tar- the extended identifiers. These 29-bit long identifiers are
geted automotive industry, but also from other areas, most needed to map predefined identifier sets for open archi-
prominently industrial automation. Even before the first tectures (e.g. SAE J1939). This concept allows nodes to
automotive CAN application reached the market (five CAN be added or replaced in a network without needing to
nodes in the Mercedes-Benz S-Class of 1990, W140), change the setup of the remaining nodes. The CAN con-
CAN was used in several industrial control networks. At troller Intel 82527 (1992) was the first implementation of
first, they were implemented using proprietary higher-layer CAN 2.0. In the beginning, some other protocols were
protocols, but soon the first standards appeared, like CAN considered for vehicle networks, but those did not go into
Kingdom, Devicenet, and SDS (Smart Distributed System). production (Abus) or they were phased out again (J1850,

Freely pro-
grammable central control unit

B
*:*0341*28
;.8-
 $
&2)
C>
B
&6,*
7;.8(-.2,
(&4&(.8=
94
83


B
0*<.'.0.8=
8-639,-
1908.+92(8.32


%7
B
<8*27.:*
(31192.(&8.32
.28*6+&(*7
B
9.8&'0*
+36
639,-
*2:.6321*287
B
8&68*6.8
+36
*&7=
&2)
7.140*
7*894

Exhibition Dates
Hannover Messe Industrial
Automation
&23:*6

@
 

Sensor + Test

96*1'*6,


@
  

Sensor-Technik Wiedemann GmbH A
1
D6*2;&0)

A
 
&9+'*96*2
A
*61&2=
A
!*0*4-32*






28*62*8
;;;7*27368*(-2./(31
 The higher bit-rate is still used in so-called "Private CAN"
25th anniversary

networks: point-to-point connections between two nodes.

While the first CAN networks used discrete transistors as bus
line drivers, this was soon followed by dedicated CAN trans-
ceiver ICs. Different transceiver types were developed, but
ISO 11898-2:2016 is now the standard in automotive applica-
tions, with or without the partial networking extension where
the transceiver is able to decode wakeup messages. In indus-
trial applications, galvanically isolated CAN transceivers are
preferred.
Besides the first two fields of application, several other
industries specified their own CAN standards, like Arinc-825
for aviation and ESA's ECSS-E-ST-50-15C for on-board
spacecraft communications and control systems. This
enabled CAN networks to reach the Moon (Smart-1, 2004)
and Mars (ExoMars, 2016). In the automotive world, CAN
remains the predominant bus system, although some other
protocols have been added for specific applications, like LIN
for master-slave sub-networks and Most for infotainment. Eth-
ernet was also introduced into automotive applications, first
only for tasks with high data volumes, but lately also for con-
trol applications.
One specific automotive protocol, Flexray, has been
added to CAN networks. Flexray was developed for x-by-
wire systems, where mechanical and hydraulic linkages are
replaced by bus systems, so that the car can be controlled
with a joystick. These control loops require synchronized
nodes communicating on redundant channels. Flexray was
developed by the Flexray consortium targeted for a bit-rate
Figure 4: Intel 82526 advertisement (Photo: Bosch) of up to 10 Mbit/s. As with CAN, the network topology needs
careful consideration to achieve high bit-rates; Flexray at 10
VAN). Ten years after the first CAN network, few cars did Mbit/s requires active stars or linear bus lines with a limited
not use CAN, and most had several CAN networks for number of nodes. A time-triggered extension of the CAN pro-
different functions. tocol (TTCAN) was developed for the same x-by-wire applica-
When CAN entered automotive volume produc- tions. TTCAN has been standardized as ISO 11898-4:2004.
tion, it turned out that the structure of the car's wiring har- The targeted x-by-wire systems never appeared in volume
ness did not agree with the original idea of how to build a produced cars, so time-triggered communication lost its main
CAN line: the so-called ISO-topology with terminations at purpose and the ECUs still operate with event-driven control
both ends of the bus line and all nodes connected by short messages.
stubs. A wiring harness is usually pre-produced in an "E"-or The ever increasing need for bandwidth in automotive
"H"-shape, so most CAN networks are built as passive control networks raised the number of CAN networks in a
star networks. This degrades the signal quality, which car, requiring gateways and a backbone for interconnection.
is why most automotive CAN networks use a bit-rate of While Flexray is already used as such a backbone, its time-
0,5 Mbit/s instead of the 1 Mbit/s for which the CAN protocol triggered communication schedule is not easily integrated into
was designed. At the lower speed, the robust CAN protocol the event-driven operation of automotive ECUs. A solution for
can easily tolerate the ringing introduced by the bus topology. this data bottleneck was required, ideally without the need of

Figure 5: The 82526 evaluation board (Photo: Bosch) Figure 6: Intel 82527 advertisement (Photo: Bosch)

10 CAN Newsletter 1/2017

 a radical transformation of automotive networking. For that
reason, Bosch upgraded the CAN protocol to CAN FD.
CAN FD was first published in a white paper in
April 2011, soon followed by hardware demonstrators. It
increases CAN's data rate in two ways: First by switching to
a higher bit-rate for the payload of a CAN frame and second
by extending the frame's data field to improve the header/
payload ratio. The maximum data length in CAN FD is 64
bytes. The rest of the Classical CAN features, like arbitra-
tion, acknowledge, and error handling, are left unchanged.
There is no disruptive transition when switching from CAN Figure 7: CAN FD frame (Photo: Bosch)
to CAN FD; the design environment and hardware can be
upgraded incrementally. CAN FD nodes are still able to For industrial automation, CiA's working groups upgrade
take part in Classical CAN network communication. the CANopen software standards to integrate CAN FD.
The physical layer and bus topology of Classical CAN Other standards (CiA 601) provide guidelines for hardware
may be left unchanged when the network is upgraded to integration. New security concepts for CAN networks have
CAN FD; the signal delay times that limit the Classical also been developed, enabled by the longer data fields
CAN's bit-rate are not relevant for the bit-rate in CAN FD's of CAN FD that allow adding a signature for message
data phase. This bit-rate is still limited by transceiver asym- authentication.
metry and by ringing on the bus line. The advantages of
CAN FD have been accepted in the CAN community, so
several CAN controller ICs are available today, as well as
design and measurement tools. Standardization started
early and culminated in the integration of the CAN FD frame Author
format in ISO 11898-1:2015. The physical layer standard-
ization in ISO 11898-2 also considers CAN FD, e.g. by new Florian Hartwich
parameters for bit symmetry. Transceivers released for Robert Bosch
CAN FD are already in production. Volume production of florian.hartwich@de.bosch.com
the first cars using CAN FD is expected to ramp up in 2017. www.bosch.com

Display Family

B
"&6.&287
+631

83

.2(-*7

B
 59&0.?
*)
-&6);&6*
&2)
73+8;&6*

40&8+361
B
<8*2)&'0*
8-639,-
*<4&27.32
'3&6)7
B
.,-
:.*;.2,
&2,0*
B
6&4-.(&0
)*7.,2
&2)
463,6&11.2,
8330

Exhibition Dates
Hannover Messe Industrial
Automation
&23:*6

@
 

Sensor + Test

96*1'*6,


@
  

Sensor-Technik Wiedemann GmbH A
1
D6*2;&0)

A
 
&9+'*96*2
A
*61&2=
A
!*0*4-32*






28*62*8
;;;7*27368*(-2./(31