Advertisement
Product Releases
Advertisement

ATCA Power Design

Thu, 06/30/2005 - 8:25am

ATCA power systems offer a number of advantages over traditional – 48 V power rails. Among them are higher reliability, open specifications and individual circuit packs.

Since the completion of the PICMG® 3.0 specification two years ago, the ATCA systems built on this specification have seen a rapid adoption by OEMs, TEMs and carriers. Some TEMs and carriers are now explicitly calling out ATCA architectures as strong preferences or requirements in defining their newer network elements.

Glossary of Acronyms

ATCA— Advanced Telecom Computing Architecture (AdvancedTCA®)

BOM— Bill of Materials

SELV— Safety Extra Low Voltage

DSLAM— Digital Subscriber Line Access Multiplexer

EMC— Electromagnetic Compatibility

ETSI— European Telecommunications Standards Institute

FET— Field-Effect Transistor

J— Joule MLBF— Make Last, Break First

OEM— Original Equipment Manufacturer

TEM— Telecommunications Equipment Manufacturer

TNV— Telecom Network Voltage



Most of the initial adoptions of ATCA systems have been in the wireless market space. Today, numerous field trials and even field deployments are underway with carrier customers transparently using ATCA-based infrastructure to communicate. There is also strong interest in ATCA meeting the fiber optic, DSLAM, application server and database elements of communications systems.


Why ATCA?

What is driving this demand? OEMs, TEMs and carriers want reliable systems that are based on open specifications. The open specifications protect against unwanted lock-in to any single vendor or supplier. Reliability is driven by the collective input of over one hundred companies that participated in the development of the PICMG 3.0 specifications on which ATCA systems are based.



click the image to enlarge

Figure 1. Typical shelf power distribution circuit.

A key element of this reliability is the power subsystem used in ATCA products. ATCA blades support redundant – 48 VDC feeds to each blade. This is in keeping with the preferred power delivery system for carriers, who prefer that individual "circuit packs" like ATCA blades receive their power directly off the – 48 V feeds rather than having shared power supplies convert from – 48 V to intermediate voltages such as 12, 5 or 3.3 V. This reduces the resistive losses, higher current requirements and inductive impedance limitations commonly associated with lower voltage power distribution schemes. The typical power distribution to an ATCA shelf is shown in Figure 1.


Negative 48 V Power

Though the input power is generically referred to as – 48 V power, the actual voltage level defined in the PICMG 3.0 specification is actually of a wider range. Most telecom systems have a nominal voltage of – 48 V, but some systems in Europe and Asia are based on – 60 V as the nominal voltage. The typical operating range of – 48 V systems is between – 40 and – 57.6 V (with excursions up to – 60 V), while the typical operating range for – 60 V systems is between – 45 and – 72 V (with excursions up to – 75 V). Overlaying these two sets of criteria gives an operating range of – 40 to – 72 V. The PICMG 3.0 specification requires all shelves to support a voltage range from – 40.5 to – 72 V, in accordance with the limitations specified in ETSI 300 132-2. However, carriers often require voltages as low as 㪿 V, measured at the input to the shelf. Since the PICMG 3.0 specification allows for as much as a 1 V drop through a typical shelf's power entry module and backplane, it is a good practice for ATCA blades to support – 38 to – 72 V incoming power.


Note that voltages between – 60 and – 80 V are classified as TNV-2 hazardous voltages, rather than the SELV ratings found between 0 and – 60 V. TNV-2 signals require special consideration when designing blades, particularly in maintaining the proper creepage (surface distance) and clearance (air gap) spacing from other conductive elements.


Typical ATCA Blade Power Subsystem

An ATCA blade's power subsystem typically comprises the components as shown in Figure 2.



click the image to enlarge

Figure 2. Typical ATCA blade power conversion.

Each blade receives two independent power feeds, typically referred to as A and B feeds, through the P10 power connector. The A and B – 48 V feeds are diode OR'd together, and the returns are also diode OR'd together. This allows the blade to continue operating without interruption if either field fails for any reason.


The PICMG 3.0 specification provides for long EARLY_A and EARLY_B contacts to pre-charge capacitors and other elements of the power subsystem before full power is applied. This is done to minimize inrush current during board insertion, and these contacts must be current limited with a resistor. However, almost all blades are going to need some form of hot swap controller to properly control the inrush current when payload power is applied. As such, the EARLY_A and EARLY_B pins should generally not be used. The commercially available hot swap controllers provide better inrush protection than the early pins and also work even when the board is not being physically inserted, such as when power is applied to a shelf with boards already installed.


The enable pins are the pins on an ATCA blade with the shortest wipe length. They are low current pins that are tied into the corresponding return feed on the backplane. These MLBF enable pins are used to tell the hot swap controller when the blade is fully inserted in the chassis. By the time these pins mate, all other power contacts are guaranteed to be mated, so the blade can start using power. Conversely, if a maintenance person yanks a board out of a shelf without following the graceful shutdown process, the enable pins will immediately signal the hot swap controller that the blade is not fully seated, and the hot swap controller can stop drawing power from the backplane (often in just a few microseconds). By the time the longer power contacts disengage, power is no longer flowing through the power pins, eliminating any possible damage from arcing power contacts.


As mentioned earlier, a hot swap controller provides a very important function in controlling the inrush current for a blade so that adjacent blades do not see a fluctuation in the input power feed when a blade is inserted. Another critical function of the hot swap controller in ATCA systems is the controller's ability to turn on or off the payload power to a blade.


Power Up!

When an ATCA blade is first inserted into a chassis, the blade does not immediately power on. Instead, a small microcontroller on the blade is allowed to draw up to 10 W of power (including power conversion) to communicate with the Shelf Manager and get permission to power on. The Shelf Manager may immediately grant the blade access to power on, or it may wait years before giving the blade the approval to turn on. Though not shown in Figure 2, some hot swap controllers provide a separate contact for a small amount of power to be drawn for system management, independent of the main FET shown in the figure. The separate management power is typically routed to a small power converter that can power the microcontroller and its associated circuitry, typically less than 1 W output power draw.


Most modern high-efficiency DC/DC power converters operate at high frequencies and can generate substantial noise on the power feeds, so all blades need to have an EMC filter to minimize emissions. To ensure worldwide opportunity for their products, blade vendors should target Class B EMC emissions levels.


The use of filter capacitors, high voltage transient suppressors and bulk capacitors helps blade designers provide a smoother input voltage to the DC/DC voltage converters on each board. Though many board designs will convert the wide-ranging – 48 V input voltage to an isolated 12 V rail and then perform secondary conversions from the 12 V rail, there are some applications where it may be more practical or more efficient to go directly from – 48 V to an isolated lower voltage rail like 2.5 V. There may also be some situations where it is prudent to have a mix of direct and indirect power conversion from – 48 V.


It should be noted that the PICMG 3.0 specification does not mandate that blades be used with OR-ing diodes (or OR-ing FETs), but that is generally the more practical approach. Approaches that use redundant power converters typically take substantially more board real estate for the converters themselves. Adding in duplicate hot swap controllers, high voltage transient suppressors, bulk capacitors, filter capacitors and other associated elements usually makes such approaches less desirable from a cost and real estate perspective. This is only compounded when one looks at duplicating the low voltage transient solutions outlined below.


Blade-level Low Voltage Transients

The under-voltage requirement in the PICMG 3.0 specification is a key ATCA feature that was driven by stringent carrier requirements. Carriers mandate that equipment meet the following under-voltage requirements:

• Voltage: 0 V • Duration: 5 ms • Fall rate: 50 V/ms • Rise rate: 12.5 V/ms


If the minimum voltage level at which the on-board voltage converters will work is – 36 V, then the total time between – 36 and 0 V is 8.6 ms. Keeping a 200 W blade running for almost 9 ms is a daunting task. Since 1 J is equivalent to 1 watt-second, a 200 W blade needs to store 1.8 J (200 W × .009 seconds) to keep functioning properly.


The typical way to store energy for an application such as this is to use one or more capacitors. The stored energy in a capacitor can be expressed by the following equation:

E = ½CV2

Where:

E is the energy in Joules

C is the capacitance in Farads

V is the voltage

A simple approach to energy storage would require significant real estate to meet this under-voltage ride-through requirement, particularly considering the voltage loss as power goes through the shelf's power entry modules, the backplane, the blade's OR-ing diodes and any high voltage transient suppression. A – 40 V input to the shelf could be – 38 V by the time it gets to the voltage converter input.


At – 38 V, a fairly large device like a 680 µF capacitor would store about 0.49 J of energy (½ × .000680 × 382). The typical cutoff voltage for most voltage converters is – 36 V. At – 36 V, the capacitor would still have 0.44 J of stored energy, meaning that only 0.05 J would be usable despite using a capacitor that is about 1 inches in diameter. This approach would be impractical to reach the 1.8 J requirement listed above.


Rather than stopping at – 36 V, a boost circuit can be added to be able to draw more of the stored energy out of the capacitor. If the boost circuit can utilize the capacitor energy down to the 5 V level, it will leave less than 1/100th J in the capacitor. However, the reduced efficiency of the boost converter will typically allow less than 0.4 J to be recovered from the capacitor. This would mean that five large capacitors like this would be required to meet the ride-through requirement.


Energy Storage Secrets

The key to understanding a solution that requires less real estate is to understand what the energy storage is when the incoming voltage is – 72 V. At – 72 V, the energy storage in that same 680 µF capacitor is 1.76 J.


If the circuit to charge the ride-through capacitor always charges it to – 72 V regardless of the incoming voltage, meeting the 9 ms ride-through requirement becomes much simpler. Two approaches emerge:

• Provide 930 µF total capacitance (such as with 680 µF +250 µF capacitors). At – 72 V, the total stored energy is 2.4 J. At – 36 V, the total stored energy is 0.6 J. This gives 1.8 J of energy to power a 200 W blade for 9 ms. • Provide 780 µF total capacitance (such as with 680 µF +100 µF capacitors) along with the boost circuit approach to utilize the capacitor down to – 5 V. The stored energy is over 2 J at – 72 V, 0.62 J at – 40 V and less than 0.01 J at – 5 V. Accounting for a 20% efficiency drop below – 40 V due to the boost circuitry, this gives almost 1.89 J of total usable energy, which is enough to power a blade for more than 9 ms at 200 W.


Different designers may choose different options, depending on whether the low voltage boost circuitry takes more space or is more expensive in terms of real estate or BOM cost than the extra capacitance shown in the first option.


Alternative Energy

Another approach sometimes used to determining the energy storage needed is provided by rewriting the earlier equation to:

C = 2 E/V2

If the system cannot use all the stored energy in the capacitor, the following applies:

C = 2 E/(V12– V22)

Where:

E is the energy in Joules

C is the capacitance in Farads

V1 is the initial (charged) voltage

V2 is the final (discharged) voltage

If it is assumed that V1 is 72 V and V2 is 36 V and that the desired ride-through time is 10 ms, the following applies:


click the image to enlarge


The net result is that blades implementing such an approach would need approximately 5.15 µF per W of power draw.


Note: For maximum reliability, capacitors should be rated for at least twice their nominal input voltage. Thus, the ride-through capacitors above should be rated for at least 145 V.


Fusing and Fault Protection

Even the best designed systems may have an electrical short at one time or another due to service issues, component quality issues or manufacturing defects. The normal protection against such faults is to include fuses to interrupt current flow in such conditions, as shown in Figure 2.


If one looks at what happens to the power feed in a system when an electrical short occurs on a blade, it can be seen that the voltage (pressure) drops as soon as the fault occurs, since there is now an unimpeded path for the electricity to flow. Depending on the type of fuse used, the fault time of the fuse may vary. Once the fuse blows, the voltage level will fluctuate briefly until the normal power delivery is established (see Figure 3).



click the image to enlarge

Figure 3. Voltage and current through electrical short.

In many systems such a fuse failure would cause other blades in the chassis to reboot due to the low voltage condition, but ATCA systems are different: One of the ATCA system's advantages is that they should be able to survive a dead short on a blade and keep on functioning properly. The fuse requirements outlined in the PICMG 3.0 specification are correlated to the low voltage transient requirements discussed earlier.


In creating the PICMG 3.0 specification, the specification developers simulated multiple different power delivery systems to analyze what effect different fuse characteristics had on the backplane voltage level. Surprisingly, "fast blow" or "slow blow" characteristics of fuses were less important than their I2t ratings. The I2t rating for a fuse is an indication of how quickly a fuse will heat up and blow. It was found that fuses with I2t ratings of 100 or less will blow in less than four seconds in even the most sensitive power environment simulated. Allowing a few ms for power to stabilize after the fuse blows still keeps the full extent of this incident to less time than the 9 ms that blades need to meet to cover the low voltage transient requirement. Net result: ATCA blades should be able to keep operating without incident, even if an adjacent blade develops a dead short.


It is important to note that all the backplane power connections have fuses on them, even the return signals. This is an often overlooked aspect of power system design that carriers require products to provide. The fuse on the – 48 V feeds is understandable since it prevents a short on the board from damaging the rest of the system or creating a fire. Similarly, the early power feeds (if used) also need fusing to ensure that they do not become a sneak path for a short to continue drawing power.


Fusing Return Feeds

Consider what happens if there is a 3 V differential between the two return signals for battery plants A and B (as may be seen in larger central offices). If a diode in the return path fails, this can short the A and B return paths together. In this condition, it is possible for all the return current to flow through this less resistive path to ground. All the return current from neighboring blades, shelves, and racks could end up being diverted through this one blade that ties the (otherwise) isolated return paths together. This could damage the blade and the shelf, possibly even leading to fire if left unchecked. By inserting a proper fuse in the return path, designers can guarantee that the fuse will blow before the shelf or other equipment is damaged.


Designers should ensure that the fuse in the return path is rated higher than the fuse in the – 48 V paths so that normal shorts blow the fuse in the input path first, keeping the return path fuse to protect against failures in the back end, such as the diode failure mentioned above.


Enable pins are tied to the return paths and need to be fused as well to ensure there is no sneak path that could circumvent the protection in the return signals as noted above.


Conclusion

The PICMG 3.0 specification provides a solid foundation on which to build ATCA systems. Each blade receives redundant power feeds for maximum reliability. Diode OR-ing and proper fusing of both the – 48 V and return feeds allows for a simplified power design that minimizes real estate impact while providing the necessary protection for deployment in carrier grade environments.


An ATCA blade's ability to survive carrier-mandated low voltage transients enables the blade to continue operating even if a neighboring board develops a short circuit. The key is in boosting the voltage to the holdover capacitor to the upper limit of the DC/DC converters. This will dramatically increase the blade's energy storage without taking much more board real estate.


Carriers today are beginning to specifically ask for ATCA solutions in their new network element deployments. Understanding the power system for ATCA blades is a key prerequisite for equipment designers to meet these customer needs.


Note: PICMG and AdvancedTCA are the registered trademarks of the PCI Industrial Computers Manufacturers Group. ATCA is a trademark of the PCI Industrial Computers Manufacturers Group. Other names and brands may be claimed as the property of others.


About the Author

Kevin W. Bross is a system architect and technical marketing engineer in the Modular Communications Platform Division in the Intel Communications Group. He has held a variety of engineering and marketing roles over his 16+ years at Intel. Kevin was a key contributor to several portions of the PICMG 3.0 specification and has been instrumental in the development of Intel's AdvancedTCA SBC, chassis, and management products. Kevin earned a Bachelor of Science degree in computer science and in mathematics along with a Bachelor of Arts degree in business administration from Principia College.

By Kevin Bross

Advertisement

Share this Story

X
You may login with either your assigned username or your e-mail address.
The password field is case sensitive.
Loading