data communications issues for digital power management 20051013

7/28/2019 Data Communications Issues for Digital Power Management 20051013

1/53

1

Data Communications I

ForDigital Power System Man

20 September 2005

Robert V. White

Artesyn Technologies

Westminster, Colorado

David L

Texas In

Dallas


2/53

2

Presentation Overview

Requirements

Physical And Fiscal

Data Flow

Characteristics Types Of Buses Issues And Constraints

Recommendations

By Data Bus By Application


3/53

3

Fundamental Requirements

Low Cost, Low Cost & Low Cost

Component - Low Cost

Development - Low Cost

Robust Carry Data Without Corruption Or InteIn The Presence Of Noise

Does Not Pass the Burden To The Ho

Must Support Time Critical Commu Address The Need For Alarms And Ale

Address The Need For Fast Host Inter


4/53

4

Additional Requirements:

Who Talks To Whom?

PMBus

PMBusInterface

IC

POL w/

PMBus

POL w/

PMBus

STD

POL

STD

POL

Local/On-Board

Power Bus

Power/

System

Interface

POL w/

PMBusBus

Converter

PMBus

PMBus

Interface

IC

POL w/

PMBus

POL w/

PMBus

STD

POL

STD

POL

V

V

V

Local/On-Board

Power Bus

Power/

System

Interface

POL w/

PMBusBus

Converter

System

Power Bus

SystemHost

Local

Area

Comm

Bus

PMBus

Interface

IC

POL w/

PMBus

POL w/

PMBus

STD

POL

STD

POL

VOUT

VOUT

VOUT

Ana

Lin

En

Pow

Local/On-Board

Power Bus

Board Level

Maintenance

Processor

POL w/

PMBusBus

Converter

System

Maintenance

Comm Bus


5/53

5


Real Time Data

Time Critical Information Of Two Ty

Events

Parametric

Fault Events Can Be CatastrophicAnd Must Be Transmitted With Mini

Parametric Data Requires Data RatOf Tens Of Megabits Per Second

Recommendation Events: Dedicated Signal Lines

Parametric: Dedicated, Customized Bu


6/53

6


Hot Swap

A Common Requirement

Does Not Interrupt Bus Traffic

Does Not Require Complex System Re

More On This Later


7/53

7

Data Communication Characteristic

Connectivity: Point-to-Point


8/53

8


Connectivity: Point-to-Point

Device A

Device B

Mux

Device C


9/53

9


Connectivity: Multi-Drop

Device A Device

Device

Device

Single Master

Or

Multi-Master


10/53

10


Directing Communication On Multi

Chip/Device Select Lines

Addressing

Hard Versus Soft Addresses

Allowable Addresses Assuring Unique Addresses

Address Ties To Physical Location Or

Address Pins Not Just Binary

Tri-State Resistor Programmable


11/53

11


Bus Contention

Bus Contention In Multi-Drop BusesUnavoidable For Multi-Master

Lossless, Bitwise Arbitration Comm

Simultaneous Attempts To Transmi Adding Priority To Messages (Like

Does Not Prevent Delayed Messag

If Bus Is Busy, Even High Priority Mes

Have To Wait Until The Bus Is Clear Continuous Stream Of Higher Priority

Can Indefinitely Delay A Lower Priority


12/53

12


Speed And Timing

Megabits Per Second Is Not The W

How Fast Can Data Get FromSender To Receiver?

Over Communication Bus, Not Fast EnFor Most Time Critical Events

Packet Overhead Reduces Effectiv

Time Critical Data Should Be Route

Dedicated Buses Between Only ThInvolved

Example: Real Time Digital Current Sh


13/53

13


Polling And Interrupts

Polling

Simpler To Implement

Detection Of Failed Or Removed Devic

Consumes A Lot Of Resources

Possible Delay Time = Refresh Rate

Alert Or Interrupt Driven

More Complicated Code

Reduces Burden On Host Quick Notification Of Events

Good Choice: Blend The Two


14/53

14


Range And Number Of Devices

Range

Often Capacitance Limited

Short Range, Open Drain Drivers = Lo

Longer Range Requires More Robust

Number Of Devices

Like Range, Often Load Limited

May Be Address Limited

Generally Not A Problem


15/53

15



Range




Number Of Devices



Generally Not A Problem

Watch Out

For The

Edges!


16/53

16



Range




Number Of Devices



Generally Not A Problem Obey TOr

Watch Out

For The

Edges!


17/53

17



Bus

Master

SlaveDevice SlaveDevice SlaveDevice

Slave

Device

Slave

Device

Repeater

SELECTOR

Switch Switch Switch


18/53

18



Bus

Master

Slave

Device

Slave

Device

Slave

Device

Slave

Device

Slave

Device

Repeater

SELECTOR

Switch Switch Switch

A Number Of Options

Are Available ForExtending Buses

And Number Of Devices


19/53

19


Clock

Synchronous Clock Signal Sent With Data

Receiving Devices Do Not Need An O

Range Limited

Asynchronous No Clock Signal Sent With Data

Each Device Needs Its Own Oscillator

Can Loose Sync On Long Strings

Of Ones Or Zeroes Bit Stuffing

Fancy Coding


20/53

20


Single Ended Or Differential Signal

Single Ended Signaling

One Wire For Data

Lower Cost, Less Complicated

More Susceptible To Noise Then Diffe

Differential Signaling

Two Wires For Data Opposite Polarity Signals On Each

More Immune To Noise More Immune To Ground Voltage Diffe

Higher Cost


21/53

21


Transmission Control And Protoco

Transmission Control Issues

Device Is Busy And Cannot Be InterruRespond To Another Request

Device Cannot Accept Data At Curren

Devices Buffer Is Full

Bus Is Busy And Device Must Wait

Protocols

Read/Write Like Message Based


22/53

22


Error Detection And Correction

Possible Errors Bit Value Changed

Beginning Or End Of A Byte Or Bit SeNot Recognized

Too Many/Too Few Bits In A Frame O Start Or End Of A Packet Or Message

Recognized

Error Detection: Parity Bit, Checksu

Error Correction More Complex

Lots More Bits


23/53

23


Fault Tolerance


24/53

24


Fault Tolerance

True Fault Tolerance

Only With

Redundant BusesAnd Transmitters And

Receivers!


25/53

25


Hot Swap

Hot Swap Issues

Removal And Insertion Without Disrup

How Does System Know If A UnitHas Been Added Or Removed?

Most Buses Support Hot Swap Fair

Implementation Issues

Making Ground Connection Last Brea

Preventing Unpowered Devices From During Insertion Or Removal

MODULE_PRESENT Signal To Assist


26/53

26


Hardware Implementation

Software Emulation Using GPIO

Possible To Do

A Source Of Many, Many Headaches

Timing Is Very Difficult Even For Slow

Integrated Solutions

Many Low Cost Microcontrollers HaveBus Interfaces Built In

Must Have For Complex Buses LikeCAN Bus And Ethernet


27/53

27


IP Issues

Standard: De Facto vs. De Jure

Who Controls?

An Organization

Single Company No One

Organization Ownership Preferred

Adopters Agreements

Compliance Assurance Royalty Free Or Not?


28/53

28

Recommendation By Bus Type

RS-232 & RS-485

RS-232 Advantages

Common Peripheral

Simple

Relatively Low Cost

Disadvantages Point-To-Point

Oscillator

Speed

Recommended Simple Point-To-Point WithLogic Level Interface

RS-485 Advantage

Different

Long DisCommun

Disadvant Additiona

$1.50 In

All Proto

Recomme Longer R

CommunRack-To


29/53

29


IC And SMBus

IC

Advantages

Common Peripheral

Simple

Very Low Cost Disadvantages

Noise Sensitivity

Bus CapacitanceLimitation

Recommended SMBus Is Better Choice In

Almost Any Case

SMBus

Advantage

Low Cos

More Ro

Additiona Disadvant

Bus CapLimitatio

Recomme On-Boar

Commun


30/53

30


SPI Bus And Dallas 1-Wire

SPI Bus

Advantages

Simple

Chip Select Lines Eliminate

Addressing Concerns Good Speed (1 MHz)

Disadvantages No Standard

Chip Select Lines

Recommended Local Interconnect Of A

Couple Of Peripherals

Dallas 1-Wire

Advantage

1-Wire

Unique I

Low Pow Disadvant

Noise Se

Proprieta

Cost

Recomme Only To

Legacy S


31/53

31


CAN Bus & LIN Bus

CAN Bus

Advantages Differential Signaling

Noise Immunity

Fault Tolerance Disadvantages Cost

Requires IntegratedPeripheral

Recommended Longer RangeCommunication Such AsRack-To-Rack And Beyond

LIN Bus

Advantage Single W

ReasonaImmunity

Disadvant Cost

Slow Sp

Complex

Recomme Not RecoSMBus ORS-485


32/53

32


USB & Ethernet

USB

Advantages

Well Supported

Hot Swap Friendly

Disadvantages Requires Hub To Initiate

All Communication

Relatively ComplexSoftware And Hardware

Recommended PC To Power System

Interface For Service

Ethernet

Advantage

Long Ha

Internet

Disadvant Cost And

Very Lar

Software

Recomme InterfaceWeb SerSystem M


33/53

33

Recommendation By Applicati

On-Board/Single Board Power Syst

SMBus

Shelf-Level/Chassis-Level Power S

SMBus If Capacitance Allows RS-485 If Not

Shelf-To-Shelf Or Rack-To-Rack

RS-485

CAN Bus


34/53

34

Recommendation By Applicati

Facility Level

RS-485 Or CAN Bus

Campus Level

Ethernet PC To Power System Manager

USB


35/53

35

Summary

No One Right Bus For All Power Comm The Scope and Benefits Extend To Beyond Th

Several Well Established Buses To Choose F

Know Your Application And Its Requireme

Be Knowledgeable About Your Choices Choose The Right Tool For The Job

Be Smart In Your System Design

Imitate Successful Designs

Understand The Constraints Before You Start

Follow The Specification!


36/53

36


37/53

Data Communications Issues For Power System Management

Robert V. White Dave FreemanArtesyn Technologies Texas InstrumentsWestminster, Colorado Dallas, Texas

[email protected] [email protected]

Page 1 of 17

AbstractDigital power management, the ability toremotely manage and configure ac-dc power

supplies and dc-dc converters, is being adopted at a

rapid rate. There are many choices to be made

when implementing a digital power management

system [1][2]. The choice of transport (data link

and physical layers) is very important. Choosing

the wrong transport for the application can create

endless problems with the system. This paper starts

with a brief history of digital power management.

It then reviews the requirements for data

communications for digital power management.

Knowing the requirements, several of the more

common implementation issues are discussed along

with recommendations for avoiding those problems.

From there, a summary of common data

communications buses along with the advantages

and disadvantages for power system management is

given. The paper concludes with recommendations

data communications in power system management

for on-board power systems, within a single chassis,

for inter-chassis communication and for facility or

campus level communication.

I. EVOLUTION OF DIGITAL POWERMANAGEMENT

Digital power management is not new. It wasbeing introduced into telephone central officepower systems in the early 1980s [3][4]. Indeed,digital power management was even being usedover wide areas using the telephone network asthe communications means [5]. By the late 1980s,even computer companies such as the DigitalEquipment Corporation were using digital powermanagement in both high end systems such as theVAX 9000 and departmental servers such as the

DEC 4000 [6].By the mid 1990s, digital power management hadmade its way into desktop and even laptopcomputers [7]. In 1994, notebook computersbegan to use a standard for power managementbased on the System Management Bus (SMBus)protocol. This standard is used today to managebattery, adapter, and backlight power. During this

time, digital power management techniques werewell established in the test and measurementindustry [8] and even into the world of highenergy particle physics [9][10]. By the late 1990s,digital power management was included in theoverall system management for desktops andsmall servers [11].

The widespread adoption of the intermediate busarchitecture (IBA), which creates a local areapower system on each circuit module, has driven

power management from the system level to thecircuit board level. Starting in 2001, IC makerslike Primarion [12], Intersil [13], Volterra [14]and Summit Microelectronics [15] started offeringproducts with digital power managementfunctions. In 2004, Power-One, Inc. introducedthe Z-Series products which featured digitalpower management both at the converter andsystem level [16]. Later in 2004, a group ofpower supply and semiconductor companiesannounced an effort to develop an open standardpower system management protocol calledPMBus [17]. The PMBus specification wasmade public in March 2005 [18].

II. REQUIREMENTSWhen choosing a data communications bus forpower system management, there are manyfactors that must be considered. A bus that is anexcellent choice for one application may be aterrible choice for another. The first step tochoosing the best communications bus for a givensystem is to clearly identify the systemsrequirements. Only when the systems needs are

clear can all the tradeoffs be made that lead to thebest communications for that system.

A. Fundamental RequirementsThere are three fundamental requirements for thedata communications bus used in power systemmanagement.
mailto:[email protected]:[email protected]:[email protected]:[email protected]


38/53

Data Communication Issues For Digital Power System Management

Page 2 of 17

First, it must be very inexpensive. Although nosystem works without a power supply, the powersystem is not considered a significant value add.System OEMS are always looking for ways toreduce the cost of the power system.

Second, the communications must be robust. Thatis, it must be able to carry data without corruptioneven in the presence of electrical noise. Ideallythere is no extra burden on the system to policethe reliability of the communication.

Third, the communication must meet the need fortransmitting time critical information withoutdelay. Failures in a power system can releaseuncontrolled amounts of energy that can destroy apower converter or the loads it powers. Also,failures that drive voltages out of tolerance canresult in corrupted data. For many abnormal

events in a power system, there must be a meansto immediately command the power system tostop transferring energy to the outputs and for thesystem to stop processing data.

B. Other Requirements1. Who Talks to Whom?A good place to start when considering digitalcommunications buses is connectivity. Thesimplest power systems have a single powersupply. Its outputs provide power to the entiresystem. The entire power management functionmay take place between a system or baseboardcontroller and the power supply.

More complex power systems may have multiplepower supplies powering a distribution that isrouted to multiple boards and loads in the system.Often many of those boards and loads have dc-dcconverters that create the voltage needed by thatboard. The power system management functionmay now require communication among the mainpower supplies, the dc-dc converters and a systemcontroller.

Even more complex power management systemsmay have local power systems on each board.These individual boards may have a local powersystem manager that communicates with multipledc-dc converters on that board. In turn, there maybe a system controller that communicates with thelocal power system manager on each board aswell as the main power supplies.

In some cases a the power system, such as a 48 Vpower system providing power to multiplechassiss, may need to communicate with a datacenter control room tens of meters away.

The possibilities are essentially endless. The

point is that the system engineer must know foreach device in the power system:

With which other devices will it becommunicating,

Will it be receiving information, sendinginformation only in response to requests, orsending information without a request fromanother device,

The time criticality of the information beingtransmitted, and

How far away physically (and electrically) arethe devices with which it will becommunicating.

As for the distance question, there are a fewimportant cases:

All of the devices are on the same circuitboard, in close physical proximity and share agood common ground (on-board system),

All of the devices are in the same shelf orchassis system and share a good commonground (shelf level or sub-rack based system),

All of the devices are in the same chassis orenclosure but do not share a good commonground (chassis based system),

The devices are in different shelves orchassiss and do not share a good commonground (inter-shelf or inter-rack system), and

The devices are spread across a facility, acampus or larger area (wide area system).

This mapping of the connections and distancebetween the power converters being managed isvery important to choosing the right datacommunications bus for a given application. The

right bus for an on-board system, for example, isnot the right bus for a facility widecommunication bus and vice versa.

2. Real Time DataAnother important question is whethercommunication bus chosen for power systemmanagement will have to carry real time data. In


39/53


Page 3 of 17

a power system, real time data come in two forms:events and parametric data.

An example of an event that needs to becommunicated in real time is an output voltageout of tolerance condition. In this case, the

system needs to stop processing data with theshortest possible to delay to prevent or minimizecorruption of the data.

An example of real time parametric informationwould be used is when multiple units arepowering up at the same time. These units mightcontinuously share information about their outputvoltage so that each voltage tracks all the othersuntil each unit reaches its setpoint value. Anotherexample would units operating in parallel passinginformation about its output current during thecurrent switching cycle to the other units so that

the units can equalize their output currents andmaximize transient response.

Passing real time parametric data is a challenge.With switching frequencies routinely at hundredsof kilohertz to megahertz, passing digital data inreal time requires passing tens of megabits persecond. Data transmission rates this fast requirespecialized and possible costly hardware andextreme care in design. It is recommended thatfor now and the near future, system engineersavoid passing real time parametric data betweenpower converters or between a power converterand a power system manager.

It is also recommended that information about realtime events also not be passed over acommunication bus. As described below, thereare too many ways that a signal can be delayedwhen trying to transmit over a bus. For real timeevents that require immediate action by otherpower converters and/or the system host,dedicated signal lines, such as POWER_FAULTsignal, are recommended.

3. Hot SwapA common requirement for a power managementsystem bus is that devices be hot swappable. Hotswapping, even if the chosen data bus nominallysupports it, requires care in design to make surethat removing or inserting devices does notdisrupt the bus. Hot swapping and how it relatesto various data communications buses is discussedbelow.

III.DATA COMMUNICATIONSCHARACTERISTICS

This section reviews the key characteristics ofdata communication buses against the needs of apower management protocol.

A. Connectivity1. Point-To-Point ConnectivityOne way is for the host device to connect to onlyone power device (point-to-point). This providessome simplicity by reducing the possibly that thebus is busy when the host or the power wants tosend a message. The disadvantage is that for thehost system to communicate with to multiplepower devices requires multiple bus connectionsor a means to multiplex the communication bus.

2.

Multi-Drop ConnectivityMore commonly, there is a single bus connectingall of the elements being managed (multi-drop).This greatly reduces the number of connectionsand board space required. However, multi-drophas two notable disadvantages. The first is thatthe host must know the address of each device orhave device selection lines to signal the device thecommunication is directed at that particulardevice. The second is the potential for buscontention. When a device has information thatneeds to be transmitted, it may be prohibited from

doing so because the bus is already in use by otherdevices on the bus. Without some form of out-of-band signaling, the device will have no way ofnotifying the host that that it has information tosend. These are discussed below in more depth.

Most digital communication buses, but not all, arecapable of multi-drop operation. For example, thesimplest UART buses, RS-232 (single ended) [19]and RS-422 (differential) [20] are point-to-pointonly.

B. Directing Communication To A SpecificDevice On A Multi-Drop Bus

Most communication buses used in power systemmanagement are multi-drop. In multi-dropsystems, there must be a way that that a device toknow that a message is intended for it.

One way is to use a device select or chip selectline. In this method, there are individualconnections from the bus master to each device on


40/53


Page 4 of 17

the bus. When the bus master wants to send amessage to a device, it asserts the device selectline, notifying the device that the next message onthe bus is intended for that device. The SPI bus isan example of a bus that uses a chip select line.

The advantage to device or chip select line is thatunique identifiers (addresses) are not needed inthe devices on the bus. While removing the needfor unique addresses simplifies the devices, adevice select line from the master to each deviceis an additional cost and adds complexity to boardlayouts. These additional traces must be added inthe design to every location where the there mightbe a device on the bus. For scalable andexpandable systems, this adds cost to the entrylevel system.

The alternative to using a device select line is to

have unique identifier (address) for each deviceon the bus. While this simplifies the hardware byeliminating device select lines, it brings its ownset of problems.

How the address is set or programmed is the firstproblem. Commonly, these are set in hardware byjumpers, DIP switches or hardwired address pins.In some cases, addresses are linked to a physicallocation, such as in a backplane, and areautomatically configured by connecting a devicesaddress pins to ground or not. In someconfigurations, the high order bits of the addresscan be programmed into the power device whileallowing the low order bits to be determined bythe pin connection. In this case the number ofaddress pins can be reduced.

Soft addressing, the setting the address ofmultiple devices over the communications bus, isnot easy. The problem is that to be able to send adevice its address over the bus, it must havealready an address to which the new address canbe sent. There are schemes that allow forsoftware configurable addresses but they all rely

on a unique identifier, such as a serial number inan IC, as the initial address. Such schemes aregenerally more complex than are needed forpower system management protocols.

Given that the device addresses are set inhardware, there are two questions:

What are the allowable addresses?

What means is used to assure that all deviceshave a unique address?

The allowable addresses are a function of the datacommunication bus specification. Some busesdefine the addressing in the specification. The IC

[21] and SMBus specifications [22][23], forexample, allow seven bits for a devices address.Allowing for the addressed reserved for in thespecifications, this still leaves more than 100addresses to the user. Many other buses, such asthe RS-485 [24] and Controller Area Network(CAN) bus [25][26] leave addressing to a higherlevel of software. In practice, the number ofaddresses available on a bus is not a limitation inimplementing a digital power managementsystem.

In general, the problem of assigning unique

addresses to devices on a bus is left to the systemdesigner. One exception is the Maxim/Dallas 1-Wire bus [27] which embeds a 64 bit identifier inevery 1-Wire IC.

For the IC bus, Philips made an interestingtradeoff that both simplified and complicatedcreating unique addresses. Although the ICspecification allows for seven address bits, Philipsassumed control of the four highest order bits.The four highest order bits were a type code thatPhilips controlled. Each licensee of the IC bussubmitted their device to Philips and Philipsassigned the type code. Philips tried, with somesuccess, to assign the type codes to minimizepotential address conflicts.

Since the IC bus patents have expired in recentyears, Philips appears to no longer be controllingthe four higher order address bits. Previously,with three address bits (pins) available, therecould be up to eight of one device on one IC bus.Different types of devices co-existed fairly welldue to Philips control of the higher order addressbits. Not, with no one controlling any of the

address bits, device makers will assign them asthey please. This does mean that systemengineers using the IC or SMBus will have tochoose their devices so that:

All the devices on the bus have different typecodes (the four highest order bits) or

No more than eight devices with the sametype code are used on any one IC bus.


41/53


Page 5 of 17

It will be interesting to see how the market sortsthis out. One approach would be to make allseven address bits available to the user. However,adding seven pins to a device is unwanted costand board space. IC manufacturers are offeringdifferent ways to set the full seven bit address

without using seven pins. For example, using tri-state pins would allow setting up 125 addresseswith five pins. Another solution is to use one ormore resistors to program the address. A currentfrom inside the IC develops a voltage across theresistor. Using an A/D converter, the IC measuresthe voltage and decodes that to an address value.This technique can easily allow for eight to tenvalues per resistor with some implementationsallowing for up to 32 values. While this savesmoney on pins, it is only economical if the ICalready has an A/D converter.

This left three bits for setting a devices address.

C.Bus ContentionA multi-drop bus is like an old style party linetelephone. When a device has a message to send,there is a chance that the bus may already be inuse. If so, the device must wait until the bus is notbusy before sending its message. This is a simplefact of multi-drop buses and cannot be avoided.By choosing the bus and protocol so that theaverage utilization of the bus is low, the chancesof a device encountering a busy bus are reducedbut not eliminated.

Suppose now that more than one device attemptsto access the bus at the same time perhaps bothhave been waiting for the bus to become availableso they can send their message. There must ameans to arbitrate among the devices wantingaccess to the bus so that one gets to send itsmessage and the others wait again for the bus tobe available.

Some buses, like IC, SMBus and 1-Wire, use asimple bit-wise arbitration. As each deviceattempts to send a bit, it compares what it istransmitting with what is on the bus. Any devicesending a logic low will prevail. A device tryingto send a logic high will see that the bus is low,not high. It can therefore detect that the bus isbusy and stop trying to transmit. These buses startthere transmission with the address of thedestination so by default, so the messages sent to

the lowest addresses become by default thehighest priority messages. This method is no-destructive. Messages that lose the arbitration donot loose or destroy the message, it is simply helduntil the bus is available.

Adding a means to add a priority to a message, asthe CAN bus does, still does not guarantee that adevice will not be blocked from sending itsmessage. As long as messages of higher priorityare being sent, a device with a low prioritymessage will have to wait. Fixes to this problemhave to be made in higher levels of the software.One way is to deliberately add a minimum timeafter one device sends a message or packet beforeit is allowed to send another. While this mayallow a lower priority message to get through, itmay block a more important higher prioritymessage.

D. Speed And TimingThe question of speed always arises when talkingabout digital power system management. The keyto determining the needed speed is to recognizethat in power system management there are threekinds of events:

Non-time critical events such as configuringand monitoring a power converter operation,

Time critical events such as issuing ashutdown command in the event of an output

overvoltage condition,

Real-time control such as passing switchingcommands from a master controller to each ofthe phases in a multi-phase buck converter,and

The first two event categories are communicationsthat might be expected between a system hostservice controller and power devices.Configuration and monitoring of power devicesby a service controller does not typically dictate ahigh speed bus or critical timing. For example,

using SMBus at 100 kHz with error detection, 6bytes of information is needed to retrieve a 16 bitvalue. Adding the other setup and hold times, thiscommunications will take a minimum of 500 s.The data communication can be made moreefficient by using commands that send more dataat one time. The data may contain current,voltage, temperature, and status values, for a totalof four 16 bit values. In this case, the bus will be


42/53


Page 6 of 17

busy for about 1 ms. The number of devices onthe bus may require faster data flow in order toservice all device in the desired time interval.History has shown that once a bus is created, thenumber of devices on the bus increases over time.

Moving to the second event category, like ashutdown command, general configuration andmonitoring may establish the delay time betweendetermining that such a command should be sentand the device actually receiving the completecommand. Using the 1 ms packet command andassuming the shut-down command has no dataand therefore requires only a three byte packet,then the designer must assume that there will be adelay of about 1.2 ms before the power devicewill respond to the shut-down command. Thisdelay may be too long for many applications.

For the most time critical events, no datacommunications bus is adequate. Aside from thetransmission time, the bus may be busy when thecritical message needs to be sent. For time criticalevents, dedicated signal lines are required. Thisadds complexity to the devices, the system hostand the layout but is a price that must be paid.

The third category of event and control requiresreal-time information that needs a dedicatedcommunication channel. A command from amaster controller to stop a switching cycle cannotbe delayed because the bus is busy with a requestto read a devices serial number. For this reason,this kind of information is typically not passedover the bus used for the typical powermanagement functions of configuration, on/offcontrol and monitoring.

As alluded to earlier in this section, transfer speedis only one issue, data packet overhead is another.Small packets typical of power systemmanagement can be overwhelmed by lots ofcommunications overhead such data framesequence number and frame synchronization bits,

rendering the rate of transfer of useful informationmuch lower than the clock rate.

A recommended strategy is to make sure thatthere are degrees of autonomous operation. Forexample, if two or more power devices areinvolved with load sharing then the requiredinformation for this function should flow betweenthe devices in a manner where other

communications can not interfere and the requirebandwidth can be maintained. This means thepower communication designer may have to dealwith multiple communication busses including thesharing of analog information.

E.Polling And Interrupts

Related to the discussions above on buscontention and speed is how the system host getsinformation about the status of the other deviceson the bus. One way, polling, is for the systemhost to continually request a status report fromeach device on the bus. The advantage to this isthat system will be able to detect a failed ormissing device fairly quickly. The disadvantageis that the system host must spend a lot ofprocessor cycles generating the status requests andprocessing the results. Also, the continuous bus

traffic, and the need to poll devices on a frequentbasis means the bus speed must be fairly high toavoid long delay times.

Another way for the system host to know thestatus of the devices on the bus is for each deviceto tell the system host when its status has changed.It can do this through a signal line that is daisychained among all of the devices on the bus andconnected to an interrupt input on the host. Theonly common data communication bus thatincludes such a signal is the SMBus, which hasthe SMBALERT# signal for just this purpose.While there is a bit of complexity in generatingthe interrupt in the device and processing it in thehost, it is an overall reduction of complexity andsystem requirements over a polling system.

Actually, the recommended approach is acombination of the two. It is recommended thatthe data communication scheme include an alertor interrupt line so that devices can notify thesystem host of a change in status in a timelymanner. It is also not a bad idea if the system hostinfrequently polls the system just to make sure

that all the devices it expects to be there areindeed active and responsive. That polling can beinfrequent and does not need to be a priority taskfor the host, simplifying it and reducing itsrequirements and the speed requirements of themain data bus.


43/53


Page 7 of 17

F. Number Of DevicesAnother consideration when choosing a datacommunication bus is the number of devices thatcan be attached to the bus. The number ofaddresses available may be a limitation but

electrical characteristics are also a consideration.Buses that use open drain drivers with pull-upresistors offer advantages in terms of cost,tolerance of short circuits, and simple detection ofbus conflicts. The IC and SMBus are buses ofthis type. The disadvantage is that the totalcapacitance on the bus must be limited in order toassure that the maximum rise time specification isnot violated. Violating the maximum rise timecan cause bus devices not to recognize STARTand STOP conditions as well as corrupting data.It is important when using a bus with open drain

drivers that the bus capacitance specification isobeyed.

Capacitance on the bus comes from both the inputcapacitance of the devices attached to the bus andthe capacitance of the signal lines. This meansthat when using IC or SMBus, a system thatspans a small physical area, such as power systemon a single board, can have more devices than asystem that stretches across the backplane of rackmounted system. As a practical matter, typicalone board systems may have up to thirty deviceson the bus if care is taken with the layout. Forsystems that stretch across the backplane of a rackmounted system, practical values are in the rangeof 12 to 16 devices.

The RS-485 bus also has an electrical limit on thenumber of devices on the bus. The specificationdefines the maximum load a device can present tothe bus and required that no more than 32 suchloads be attached to any one bus. However, manydevice manufacturers provide RS-485 interfaceICs with one half, one quarter or one eighth unitloads. This enables the bus to support 64, 128 or

even 256 devices. Examples of such devices arethe SN65HVDxx series from Texas Instruments[28].

G.RangePower management systems are used overdistances ranging from centimeters to kilometers.Matching the right data communications bus tothe application is not difficult but it is important.

Using a short range bus for a long rangeapplication leads to errors and a non-functionalbus. Using a long haul bus for a short rangeapplication wastes a lot of money.

Buses that are good for short range (centimeters to

a meter or so) are IC, SMBus and SPI. Goodmedium range buses (1 meter to tens of meters)are the RS-485 and CAN buses. For hundreds ofmeters and more, Ethernet is a good choice.

If needed, the range of a bus can be extended byusing repeaters.

H. ClockWhen transmitting data on a bus, the receivingunit needs to know when a bit of data is valid onthe bus. One way to do this is for the sendingdevice to also send a clock signal along with the

data. This technique, called synchronous datatransmission, is used in buses like IC, SMBusand SPI Bus. The key advantage is that thereceiving devices do not need an oscillator, asignificant cost savings.

Synchronous mode works well when thetransmission distance is limited. For longerdistances, asynchronous mode is used. In thismode, the transmission starts with a synchronizingsignal. The receivers detect this signal and adjusttheir local clock signals to be in time with thetransmitted data. Then as the sending device putsdata on the bus, the receivers read the bus whentheir local clocks tell them it is time for anotherbit.

The advantage to this is that another signal linefor a clock signal is not needed. The maindisadvantage is that each device on the bus needsto have a precision clock. Even with precisionclocks, during long strings of the same value(either zero or one), the receiver can losesynchronization with the sender. This causes thereceiver to read the bus at the wrong times and

data errors are certain.To avoid this problem in asynchronous mode, bitstuffing is sometimes used. This means that aftera fixed number of bits of the same value are sent,a bit of the opposite value is sent just to enable thereceiver to maintain clock synchronization. Thereare also other techniques, such as return-to-zero(RZ) or Manchester coding, that are used to


44/53


Page 8 of 17

maintain synchronization. Any of thesetechniques costs in either device complexity, busbandwidth required, or both.

For power management systems, with theirtypically limited physical scope and emphasis on

low cost, synchronous buses are recommended fortheir low cost and overall simplicity.

I. Single Ended Or DifferentialData communication buses suitable for powersystem management are available with bothsingle-ended and differential signaling.

Single ended signaling, requiring only one wirefor the data, is simpler, lower cost and requiresless space on the board for routing.

The key advantage to differential signaling inpower system management is the ability to have ahigher immunity to noise. The most commonbuses with differential signaling are RS-422,RS-485, CAN bus and the Universal Serial Bus(USB) [29].

Another advantage is that most of the differentialsignaling buses work in the presence of adifference in ground potential among devicesattached to the bus. However, single ended busescan use level shifters or optical isolators toovercome differences in ground levels.

For on-board power systems, single endedsignaling is adequate. Care must be taken withlayout to minimize noise pickup, but this is aroutine matter.

For shelf or chassis level communication, singledended signaling usually works well. Noise pickupis somewhat more of a concern but good designpractice usually makes this a non-issue. IC andSMBus are routinely deployed in this applicationwith low cost and good results.

For some of the common data communicationbuses such as IC, the concern with shelf level

buses is the total capacitance. If there is asignificant length of bus traces on each board,then when several boards are plugged into abackplane the total capacitance may becomeexcessive. In this case, a repeater between the on-card bus and the traces on the backplane thatconnect all the cards in the shelf will help.

For longer distance communication, such as shelf-to-shelf or rack-to-rack, a bus with differentialsignaling is more appropriate. The ability to workwith a difference in ground potential is especiallyvaluable in this case. The most popular bus usedfor power system management in these

applications is RS-485 although the CAN bus isgrowing in popularity.

J. Transmission ControlThere are many considerations when dealing withtransmission control. Some of these are:

Device Busy the slave device maybeperforming an operation that cannot beinterrupted to respond to device request,

Device can not accept data at the currentspeed or its buffer is full, or

The bus is busy and the device must wait.In the first case listed above there are severalstrategies that can help mitigate this problem. Thestrategies depend on the type of communicationrequest. For example, the host has requested thatoutput voltage from a device. A strategy wheresuch a command requires that the device measureand then report the information may hold-up thebus so it may be better to simply report the latestmeasured value. In this way the host will alwaysreceive data that is no older than a fixed amountof time. This method may also have some issues.What happens if the data is being updated at thesame time that it is being requested? Thisassumes that the slave communication peripheralcan operate while the slave CPU is gathering data.In this case dual-ported registers can be used or asmall hold-off time is required. This isparticularly important when requesting 16 bit datafrom an 8 bit device.

In the second case, the pace of the data flow needsto be moderated to the point the slower device cankeep up. In a single master implementation,

transmission control is limited to making sure thatthe slave device is capable of accepting the data orsupplying the data. In an asynchronous bus likeRS-232, additional control lines are used or theflow control can be done by taking advantage ofthe full duplex nature to send flow controlinformation.


45/53


Page 9 of 17

In synchronous applications like IC, then clockstretching can be used to set the transmissionpace. Clock stretching has been a point ofaggravation for some system designers. Duringclock stretching, the slave device will pull theclock low while it is deciding how to respond.

This technique is used mostly during theAcknowledge bit phase. Clock stretching requiresthat master device monitor the clock line todetermine when the slave has released the linebefore proceeding. In those cases where clockstretching is used across an isolation boundary,the isolation technique must provide bidirectionalcommunication.

In the event that data is not ready, the slave devicemust inform the requesting device. However, notacknowledging (NAKing) a command is not agood solution as it may confuse the requesting thedevice as to what the issue actually is. NAKing acommand may indicate that the command is notvalid as well. Since the requesting device doesnot know which case this might be, this is aproblem. The best solution is to avoid the datanot ready issue and only use the NAK to informthe host the command is not valid.

In multi-master communication schemes, buffersand sequencing information is used to reduce theneed for flow control.

K.Data Transfer ProtocolsThere are two basic protocols for moving dataover a communication bus. In the first, the entiretransaction takes place in a single long exchangebetween the sender and the receiver. The datatransfer is very much like writing to a memorylocation for sending data or reading from amemory location when data is being received.This technique is used by the IC, SMBus and SPIbuses. This tends to be a very efficient method oftransferring data with a minimum of overhead. Itis well suited for use in local power system

management protocols where the typical amountof data being transferred is one or two bytes.

The second method of data transfer is to use amessage based protocol. In a message basedprotocol, the system host would start be sendingout a message that said Device number 3, this isthe system host calling. Send me the value ofyour output current. This message would

propagate over the bus and be received by devicenumber 3. Device number 3 would respond bysending a message on the bus that said Systemhost, this is device number 3. My output currentis 16.2 amperes. This method of communicationis best when the devices are not in close proximity

or when all of the devices on the bus areintelligent and fairly autonomous. The CAN bususes a messaging based protocol. Message basedprotocols are generally too complex and requiremore overhead than necessary for most powersystem management protocols.

The UART based protocols, RS-232 and RS-485can use either depending on how the driversoftware is written.

L. Error Detection And CorrectionThere are many possible errors in a datacommunication bus. Some are at the messagelevel and some are at the bit level. Possible errorsinclude:

A bit value being changed, The beginning or ending of a byte or other

sequence of bits not properly recognized,

Too many bits or bytes in a frame or packet,and

The beginning or ending of a message orpacket not properly recognized.

There are various methods used to detecttransmission error. In some cases, parity can beused to detect problems byte by byte. Packet ordate frame error detection is usually done usingeither checksum or CRC, Cyclic RedundancyCheck, values to indicate errors. CRC providesthe better error detection but it does add slightlymore complexity to solution.

Correction of data is certainly more complex andtypically requires sending more information suchas ECC, Error Correcting Code, in each packet so

that bit error can be corrected. The length of thecode determines the number of bits of data thatcan be corrected. For example, 8 bits of ECC candetect and correct 1 bit in 64 bits. This same codecan also detect multiple bad data bits withoutcorrection.


46/53


Page 10 of 17

M.Fault Tolerant Data CommunicationAny data communication bus can be disrupted byeither faulted signal line (open, shorted to ground,shorted to a supply voltage) or a failed device onthe bus (I/O pin open, I/O pin shorted or stuck

low, I/O pin stuck high, or an internal devicefailure that causes the device to stopcommunicating).

The only way to make the data communicationbus fault tolerant is use redundant data buses. Theredundancy must generally include a redundanthost system device if the system is to be tolerantof all single point failures. At the power devices,there can either be a multiplexer to select betweenthe two buses (lower cost) or redundanttransceivers inside the power device (higher cost).The expense and complexity to fully implement a

fully fault tolerant power system management busmeans that this is rarely done.

One of the features of the CAN bus is its ability totolerate a fault on one of the two signal lines. Inthis case the bus can continue operating withlimited speed and range.

Most faults in a power system result from a failedor malfunctioning device. This can be handled byusing a method of isolating the device from thebus. In some cases, multiplexers are used to fanout the bus to multiple devices. This method is

well suited for a single master topology.However, multiplexers add cost and complexity toany solution. If an alert response is required thenidentifying the alert device is slowed when using amultiplexer.

N.IP IssuesWhen a bus is considered to be a standard, then itusually has an organization associated with it. Inthese cases, an adopters agreement is needed sothat the end customer can have faith that thesupplier is privy to the specifications. These

buses include SMBus, PMBus, and CAN bus.

Other buses may have licensing agreements.These are typically owned by a single companyand are not supported by an industry group. TheMaxim/Dallas 1-Wire bus, for example, requirespayment of a royalty for each device on the bus.This is required whether one uses a 1-Wire device

from Maxim or if the 1-Wire protocol is emulatedin a microcontroller.

Of course, given that cost must be minimized indigital power management, royalty free buses arerecommended.

O.Hot Swap CapabilityIn many digital power management applications,it is required that units be hot-swapped. Thiscould be ac-dc front end power supplies removedfrom or inserted into a shelf with other powersupplies. Another possibility is dc-dc converterson a circuit board that is being swapped. Whenhot-swap is required, there are two key questions:

Can the devices on the unit being removed (orinserted) detach (or attach) without disturbingthe operation of the bus?

And even if a hot-swap operation issuccessful, how does the host system knowthat a unit has been removed or inserted?

Of the common data communication buses,several support hot-swap very well. RS-485,USB, CAN bus and Ethernet all have hot-swapcapability built in. With a bit of care in thedesign, the IC, SMBus and SPI buses can bemade to work with hot-swap.

The usual problem associated with hot-swappingthese buses is the loss of the bus ground return

prior to disconnecting the remainingcommunication lines. This is usually avoided byusing a longer pin for the ground connection thanfor the signal connections.

Another common issue is an unpowered devicebeing attached to the bus. In this case, there canbe a path from the signal pin through the device toits supply voltage pin which is effectively atground potential. When the device attached to thebus it effectively shorts out the bus until thedevice is active. In addition, the flow of currentthrough the device in this manner can cause lock-up so that the device never operates properly.This may mean that some form of switch, such asa small signal MOSFET, may be needed in serieswith the signal line to prevent the device fromshorting the bus when it is inserted. The switchcan be turned on and the device connected to thebus once the devices power supply is establishedand the output is well controlled.


47/53


Page 11 of 17

Extraction can also cause problems. It is best ifthe device being removed has the time togracefully disconnect itself from the bus, such asby setting tri-state outputs to be open or openingthe small signal MOSFET switch mentionedabove. Long pin/short pin arrangements are often

used to provide this notice but need carefulattention to tolerances, both electrical andmechanical, to work well. More expensive, butproviding more assurance that the bus will not bedisrupted during an extraction, is a switchattached to an extractor handle.

The other question of notifying the system hostthat a device has been removed or insertedtypically requires some form of aMODULE_PRESENT signal. This is most oftenimplemented by a pin that connects to ground.When the unit is inserted, the pin is shorted andthe host knows that module is present. When thedevice is removed the signal line is pulled highand the host knows that the device has beenremoved. This requires that the system monitoraddition lines.

For digital power management systems, hot-swapis generally not an issue. For some buses, such asIC, it is necessary to take some care in the designof each unit to make sure the bus is not disturbedduring removal or insertion. Carelessness in thisarea of design causes problems for many system

designers.

P. Hardware ImplementationDepending on the complexity of the bus, thecommunication protocol can be either integratedinto the power device controller or emulated insoftware. In some cases a portion of the protocolis implemented in hardware while the remainderis performed in software. There are many factorsto consider when making this choice. Integratedperipherals typically provide the best performancewhile decreasing the software overhead. Support

for busses like the CAN bus, USB, and Ethernetare done with integrated peripherals. Theseprotocols are considered too complex to performeverything is software.

Simpler buses like IC and SMBus are typicaltargets for software emulation, known as bitbanging. However, many system designers havehad their projects run very late due to this

decision. The attraction is usually lower cost inhardware but the specifications may be difficult tomeet without dedicating significant resources tothe protocol. For example, start bit detection usedin IC derived protocols requires an interruptservice or a very fast clock service routine. A

start bit may occur less than 5 s after the lastclock for the 100 kHz specification.

Many low cost microcontrollers, such as thosefrom TI, Microchip Technology and Atmel, havea generic clocked data peripheral that implementsthe fast service requirements of a bus whileleaving the command service to software.

IV.RECOMMENDATIONS BYBUS TYPE

A. RS-2321. AdvantagesThe RS-232 is typical implemented using aUART (Universal Asynchronous ReceiveTransmit) peripheral although software can alsobe used with general purpose I/O (GPIO) ports tobuild the interface. RS-232 provides a simple andrelatively low cost point to point communicationmethod. Typical RS-232 implementationsprovide parity for simple error detection as well asframing error detection which detects a missingstop bit. Other optional levels of sophistication

extend to flow control where device with smallbuffers can hold off data transmission until theprevious received data can be retrieved. Typicalgate count required for minimal RS-232communication is about 2000.

2. DisadvantagesDisadvantages include the limitation to point topoint communication for most applications. Inorder to access multiple power devices thecommunication lines must be either multiplexedor the host controller must have multiple interface

circuits. Either of these methods adds cost to theimplementation.

The other cost adder is an accurate oscillator. Theclock mismatch between the sending andreceiving devices should not exceed 3.3%. Mostdesigners use a 2% mismatch limit for the designspecification. This mismatch limitation restrictsthe choice of oscillator to either crystals or high


48/53


Page 12 of 17

quality resonators. The lowest cost crystal is thetype used in the real-time clock circuits. Thiscrystal oscillates at 32.768 kHz and costs less than$0.10 in high volume. The typical UART requiresa clock frequency that is 16 times the baud rate.So unless there is a circuit to multiply the clock

frequency, only 2k baud can be used.

3. RecommendationRS-232 is OK for simple point-to-pointcommunications especially when using only alogic level interface. Adding the driver for fullEIA specifications will add cost. The oscillatorsneeded in each device will add cost but a practicalsolution may be to use a common clock source.

B. RS-4851. AdvantagesThe RS-485 communication interface adds a twowire differential transceiver to the RS-232interface and adds the capability for true multi-drop communication. The RS-485 standarddefines that these transceivers can be used tointerface a total of 32 devices on the bus. Thedifferential two wire interface provides very goodnoise immunity for high speed and relatively longdistance communication. If using RS-485 withthe full EIA interface, then the specification alsoprovides some immunity to ground leveldifferences between devices on the bus.

The software driver provides for the multi-pointcommunication and defines the addressingstrategy. The simplest implementation is for asingle master bus. For long distancecommunication Category 5 cable can be used for alow cost connection.

2. DisadvantagesThe principle disadvantage is the added cost of thetransceiver. In very high quantities the price is inthe range of $1 to $1.50. The other disadvantages

are similar to the RS-232 communicationdiscussed earlier such as the oscillatorrequirements.

3. RecommendationRS-485 communications provide a solution forlong distance applications requiring good noiseimmunity and moderate speed. Good applicationsare for shelf-to-shelf, rack-to-rack

communications. RS-485 is also an acceptablechoice for longer distance communication such asbetween a power system controller in a stand-alone 48 V power system and a data center controlroom.

C.IC

1. AdvantagesIC is a simple two wire clocked data multi-dropbus. The main advantage is it low cost ofimplementation. A simple IC peripheral requiresless than 900 gates to implement. The interfacemay also be implemented using software andGPIO. This communications bus does not requireoscillator matching between devices. When anIC bus is designed to meet the IC specificationrequirements, this bus provides a good solution to

short haul communications.2. DisadvantagesThe main disadvantage is noise sensitivity. Noisecan cause data corruption and erroneous detectionof Start or Stop conditions. Bus capacitance canalso result in edge detection delay causing datacorruption and failure to detect Stop or Startconditions. The main problem with ICimplementations are due to system designs that donot meet the specifications of total buscapacitance, data hold times and start/stoprequirements.

For example, a typical I/O pin on amicrocontroller has a capacitance of about 10 pF.The maximum allowed total bus capacitance is400 pF, so no more than 40 devices can be placedon the IC bus before this specification is violated.In addition, capacitance from PCB traces mightadd about 1 pF per inch of route in a typicalapplication board which would further reduce thetotal number of devices that can be added to thebus.

Like any multi-drop bus, an active failure on a

communication line may shutdown all buscommunication.

Table 1 shows parameters that can cause problemswith Ic operation if not properly obeyed in thedesign.


49/53


Page 13 of 17

3. RecommendationThe IC bus is a very low cost solution for a localarea power management such as on-board powersystems or at the shelf level. A better choice isthe SMBus which offers the same low cost but

more robustness.Table 1. 100 kHz IC Parameters That May Cause

Issues

Parameter Min Max

Total Bus Capacitance 400 pF

Start Hold Time Before ClockGeneration

4 s

Stop Set-Up Time 4 s

Data Set-Up Time 250 ns

Clock And Data Rise Time 1 s

Clock And Data Fall Time 300 ns

D. SMBus1. AdvantagesThis standard adds to IC specification requiringstronger sink capabilities and minimum clock rate.The standard also calls for bus time-out valuesthat can help detect data line locked lowconditions. In multi-master application, theSMBus also calls out the arbitration andmaximum bus time before a bus idle conditionmust occur. The specification also includes an

optional Packet Error Checking (PEC) for errordetection. The SMBus specification requires thatall slave devices acknowledge their address. Thisrequirement is a benefit for power management asit informs the host that the device is present. Thisbus has been used for more than 10 years innotebook battery management. SMBus needsonly about 1700 gates to implement excludingaddress arbitration.

Table 2 shows improvements the SMBus madethat make it a much more robust bus than IC.

2. DisadvantagesLike the IC bus, the system must be designed tomeet the specification.

3. RecommendationThe SMBus is the best choice for local area powersystem, either on-board or at the shelf level. It isas low cost as IC bus but more robust with

timeouts, packet error checking and an interruptline (SMBALERT#).

Table 2. SMBus Timing Parameters Improve Bus

Traffic Management

Parameter Min Max

SMBus Operating Frequency 10 kHz 100 kHz

Data Hold-Time Prior To NextClock Generation

300 ns

Master Cumulative Clock LowTime

10 ms

Slave Cumulative Clock LowTime

30 ms

Maximum Clock Low Time 35 ms

E. SPI Bus1. AdvantagesThe SPI Bus is a very simple clocked data multi-drop bus. The addressing of the devices on thebus is accomplished using individual chip selectlines to each device. This reduces the circuitryrequired on each slave device which results inlower cost. The SPI bus can also be used tosupport higher data transfer speeds due to itssimple addressing structure. With the SPI bus,full duplex operation is possible. TraditionallySPI is used to support very simple devices such asexternal memory and simple peripherals. Thetypical SPI slave interface requires less than 1000

gates for implementation.

2. DisadvantagesEach device requires bus connections with a totalof N+3 bus lines. Although the chip selectaddressing mode greatly simplifies the addressingof slave devices it complicates buses that requirealert functionality.

A key disadvantage of the SPI bus is that there isno formal specification and no means ofconfirming compliance.

3. RecommendationThe SPI bus provides a simple bus that may beused to connect local devices to a host device or toextend the memory or peripherals.


50/53


Page 14 of 17

F. CAN bus1. AdvantagesCAN bus is 2 wire multi-master differential bus.The bus supports up to 1Mbit/s transfers for busesas long as 40m. The specification is allows for

high reliability transmission in very noisyenvironments. The noise and fault tolerance arethe principle benefits of the CAN bus. CAN busis available on many intermediate and high-endmicrocontrollers.

2. DisadvantagesThe CAN bus requires an oscillator with atolerance better than 1.6%. It is a messagingbased protocol and does not have a fixed means ofsetting device addresses. The data frame is largecompared to the amount of data that is typically

transferred in digital power management (16 bits)making the bus busier than perhaps necessary.The implementation size is about 4000 gates. Theimplementation is so complex that it requiresintegrated control circuitry. Emulating insoftware (bit banging) is not consideredfeasible.

3. RecommendationThe CAN Bus offers a solution for longer distanceapplications requiring good noise immunity andmoderate speed. Good applications are for shelf-

to-shelf, rack-to-rack communications. RS-485 isalso an acceptable choice for longer distancecommunication such as between a power systemcontroller in a stand-alone 48 V power system anda data center control room.

G.Local Interconnect Network (LIN) Bus1. AdvantagesLIN bus [30] is a single wire bus supportingsingle master and multiple slaves. This bus is alower cost alternative to CAN bus and providesreasonable noise immunity. It is based on thecommon UART byte interface which allows manycommon microcontrollers to emulate the LIN busin software. The LIN bus uses a checksum todetect data errors.

2. DisadvantagesThe oscillator tolerance needs to be better than 2%between master and slave during any frame

transfer. The maximum transfer speed is 20kbit/s. The bus interface is open drain and uses apull-up resistor so all of the cautionary commentsabout specification limits on bus capacitanceapply. Data frame length includes a header that islonger than other simple interface methods. The

LIN bus takes about 3000 gates to implement.

3. RecommendationThe LIN bus has no advantage over other lowercost and more available buses for power systemmanagement. The SMBus is recommended forsimpler, lower cost applications. For longer ormore noise immune data communication, useeither the RS-485 or CAN buses.

H. USB1. AdvantagesThe USB is a well supported interface withconsiderable hardware and software support.USB provides a simple interface into PCoperating systems. The USB is very good at hotswapping.

2. DisadvantagesThe USB requires a hub for communications tomultiple devices. The USB function needs about15K gates for implementation on a power device.The USB is a single master bus. Allcommunications are initiated by the hub.Attached devices are not able to independentlysignal the host system that they need attentions.For applications like power system management,this means that the host must constantly poll all ofthe attached devices.

Another disadvantage to the USB is that the use ofthe name or logo requires formal compliancetesting, an unwanted expense for most powersystem management applications.

3. RecommendationA good use for the USB in power systemmanagement is to provide a familiar interface forservice people to access power system controllers,such as those found in telephone central officebattery plants. This interface has typically beenan RS-232 serial port. As serial ports disappearfrom personal computers, these serial ports areexpected to be replaced by USB ports.


51/53


Page 15 of 17

I. Ethernet1. AdvantagesEthernet is very good for long haul and Internetconnectivity.

2.

DisadvantagesEthernet packet size is not efficient for the typicalmessage traffic expected for power management.Requires message based transactions, not memoryread-write transactions. Ethernet requires about20K gates to implement and the software supportusually is implemented using an embeddedoperating system.

3. RecommendationEthernet is recommended for wide area networkaccess to power system controllers such as those

found in central office battery plants or wirelessbase stations. It is especially recommended forthose power system controllers that are hostingembedded Web servers.

V. RECOMMENDATIONS BY APPLICATIONA. On-Board Power SystemsFor on-board power systems, the SMBus is therecommended data communications bus. It offersthe low cost of IC but with a much greaterrobustness. The second choice would be the IC

bus due to its low cost and wide availability ofdevices with integrated silicon solutions.

B. Shelf Or Chassis Level Power SystemsFor shelf or chassis level systems, such as thosemounted into 19 inch racks, the SMBus is againrecommended if the total bus capacitance allowsthe specification to be met completely.

If the SMBus is not suitable, then considerSMBus for the on board communication with abridge to RS-485 for the slot-to-slot or slot-to-hostcommunication. The second choice to the RS-485

for the slot-to-slot or slot-to-host communicationwould be the CAN bus.

C. Shelf-To-Shelf Or Rack-To-RackCommunication

For power system management protocols thatrequire communication from shelf-to-shelf or

from rack-to-rack, RS-485 is recommended. Thesecond choice would be the CAN bus.

D.Facility Or Campus Level CommunicationFor power system management within a smallfacility, where the transmission distances are 100

meters or less, either the RS-485 or CAN bus canbe considered. At these distances, using Ethernetstarts to have some advantages. For example, ifthe power system manager in the system needs anembedded Web server to provide an HTMLinterface to the facility level controller, thenEthernet would be the right choice.

E. PC to Power System ManagerCommunication

For larger power systems, such as those found in atelephone central office or even a wireless base

station site, connectivity to a personal computer isoften required. A service person might carry alaptop computer that they want to plug into thepower system to monitor status or makeconfiguration adjustments. In this case, there aretwo possible connections.

One is USB. This provides a familiar and hotpluggable interface for the service personscomputer. One disadvantage is that the computerwill need more complex software to provide theuser interface and the communication with the

power system manager.The other choice is Ethernet. If the power systemmanager includes an embedded Web server, thenthe service persons computer only needs a Webbrowser to communicate with the power system.This simplification of the personal computerrequirements often overcomes the additional costof the embedded Web server in the power systemmanager.

VI.SUMMARYThis paper has provided a comprehensive

overview of data communication issues for powersystem management protocols. The requirementsof a power system management communicationwere reviewed. The attributes of datacommunication buses were presented andanalyzed for their usefulness and appropriatenessfor power system management applications.Recommendations were then made for the best


52/53


Page 16 of 17

use of each of several common datacommunication buses. Also, for several commonpower system management applications,recommendations were made for the mostappropriate bus for those applications.

There is no one right data communication busfor all applications. It is always best to have theright tool for the job. That said, for power systemmanagement systems that are localized to a singlecircuit board, shelf level/rack mount system orthose in enclosures small enough to have a goodcommon ground, the SMBus is recommended asthe data communication bus of choice. It is aslow cost as IC but far more robust. It alsosupports features like the SMBALERT# signalwhich can be used as an interrupt line andremoves the need for continuous polling. Forlonger distance communications, such as betweensystems mounted in different rack, then either RS-485 or Can buses would be a good choice.

REFERENCES

[1] R. White, Digital Power SystemManagement, Proceedings Of The AppliedPower Electronics Conference, APEC 2006,Vol. 1 pp. 176-182

[2] R. White, Considerations For A PowerSystem Management Protocol, DarnellDigital Power Forum 2004, Presentation 5-1

[3] P. F. Gensinger and E. A. Rosin,Microprocessors in power control, the timehas come, Proceedings Of The ThirdInternational Telecommunications EnergyConference, INTELEC 1982, pp. 1-6

[4] T. V. Papathomas and R. Suderi, Storedprogram control of dc power plants,Proceedings Of The Third InternationalTelecommunications Energy Conference,INTELEC 1982, pp. 18-23

[5] W. S. Watson, Distributed energymanagement systems using the publictelephone network, Proceedings Of TheThird International TelecommunicationsEnergy Conference, INTELEC 1982, pp. 1-6

[6] Charles Butala, HP, Marlborough,Massachusetts, Private conversation

[7] Intel et al, System Management BusSpecification, Revision 1.0, September 15,1995.

[8] P. Birman and S. Nercessian, Programmablepower supplies use switch mode topologies,

Electronic Products, March 1995, pp. 33-34.[9] A. Akiyama, T. Nakamura, M. Yoshida, T.

Kubo, N. Yamamoto and T. Katoh, KEKBPower Supply Interface Controller Module,Proceedings Of The International Conferenceon Accelerator and Large ExperimentalPhysics Control Systems, ICALEPS 1997

[10]T. T. Nakamura, A. Akiyama, T. Katoh, Ta.Kubo, N. Yamamoto and M. Yoshida,Magnet power supply control system inKEKB accelerators, Proceedings Of TheInternational Conference on Accelerator andLarge Experimental Physics ControlSystems, ICALEPS 1999, pp. 406-408

[11]Intel, NEC, Hewlett-Packard and Dell, IPMI Platform Management FRU InformationStorage Definition, Revision 1.1, 27September 1999

[12]R. Zahn, Digital multi-phase power controlfor advanced processors, Intel TechnologySymposium, Power Track Presentation 9,September 2002

[13]S. Ohr, Intersil teams with Primarion topower high speed processors, EETimes, 15October 2001

[14]A. Burstein, Volterras 3rd generationintegrated switching regulators for Intelmicroprocessor power, Intel TechnologySymposium, Power Track Presentation 8,September 2002

[15]Summit Microelectronics, SummitMicroelectronics Expands PowerManagement Family With New Sequencing

Power Supply Controller for Point-of-LoadApplications, Press Release, August 1, 2002

[16]L. Pechi, Z-One digital IBA logistic andperformance benefits, Proceedings of theDigital Power Forum 2004, Paper 4-1.

[17]Digital power management protocol,Power Electronics Technology, October 1,2004


53/53


[18]http://www.pmbus.org/news.html[19]ANSI/TIA-232-F, Interface Between Data

Terminal Equipment And Data Circuit-Terminating Equipment Employing SerialBinary Data Interchange, Telephone

Industry Association, 1997[20]ANSI/TIA/EIA-422-B, Electrical

Characteristics of Balanced Voltage DigitalInterface Circuits, Telephone IndustryAssociation and Electronics IndustryAssociation, 1994

[21]The IC Bus Specification, Revision 2.1,Philips Semiconductors, January 2000

[22]System Management Bus Specification,Revision 1.1, Intel Corporation et al,December 11, 1998

[23]System Management Bus Specification,Revision 2.0, Intel Corporation et al, August3, 2000

[24]ANSI/TIA/EIA-485-A, ElectricalCharacteristics of Generators and Receivers

for Use in Balanced Digital MultipointSystems, Telephone Industry Associationand Electronics Industry Association, 1998

[25]CAN Specification, Revision 2.0, RobertBosch Gmbh, 1991

[26]ISO 11898-1, Road Vehicles - ControllerArea Network (CAN) - Part 1: Data LinkLayer And Physical Signaling, Revision 03,December 1, 2003

[27]http://www.maxim-ic.com/1-Wire.cfm[28]Kevin Gingerich, The RS-485 unit load and

maximum number of bus connections,Texas Instruments, 2004

[29] Universal Serial Bus Specification,Revision 2.0 and errata, Compaq, Hewlett-Packard, Intel, Lucent, Microsoft, NEC, andPhilips, April 27, 2000

[30]LIN Specification Package, Revision 2.0,LIN Consortium

data communications issues for digital power management 20051013

Documents