| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
This message is from Brian Lynch at TI, his pdf file was too big for the ad-hoc reflector.
So, I sent it to David Law, who kindly put it in the documents folder.
I am forwarding this email to let everyone know that Brian's file is located at:
http://www.ieee802.org/3/af/public/documents/inrush_AB.pdf
Please refer to this file along with the text below.
thanks.
- Rick
-----Original Message-----
From: Lynch, Brian
Sent: Wednesday, May 02, 2001 11:58 AM
To: stds-802-3-pwrviamdi@ieee.org
Subject: Inrush limiting - PSE or PD - Revisited
All,
Just after the last meeting in Hilton Head, there was a flurry of Email on the reflector regarding inrush current limiting, and where it should be placed: The PSE or the PD.
Since then, I have put together two simulations. The first showing operation with inrush limiting in the PSE (Method A) , and the second with inrush limiting in the PD (Method B). The first page of the attached .pdf files shows the schematic used for Method A. (The circuit for Method B is similar. I have left it out to keep the file small. I 'll send a copy if requested.)
In both cases, there is a PSE which includes a bulk power source and a switch in series to connect it to the PD after discovery. In the PD there is a DC/DC converter, separated from the PSE by a switch and an circuitry to control that switch.. The DC/DC converter is modeled as an average model to speed up simulation time. The load resistor at the far right is used to vary the power required by the PD.
I wanted to share my findings with the group before the meeting later this month in St.Louis, so that (hopefully) we can come to a conclusion. If anyone is interested in seeing more wave forms from my simulations, let me know. I have not included too many of them here because of file size limitations.
Background:
During the discovery/classification process, there is a requirement that the capacitance across the power lines be limited to a small value. This is in contrast to the need of the follow on DC/DC converter, which requires a relatively large capacitance at its input. To accommodate both requirements, a "switch" is placed in series with the input line in the PD. This switch disconnects the capacitor from the PSE during detection/classification, and connects it, and the DC/DC converter, to the PSE for normal powered operation.
When the switch is closed and the capacitor is initially connected, there is a large inrush of current into the bulk capacitor as it charges. The following is a comparison of two methods to manage this inrush current. In addition, the circuits used in each method must protect themselves and connected equipment from failure in the event of a fault occurring.
Method A: Inrush limiting in PSE.
Here, the PSE current limiting is set at a level below that in PD. This means that a PD will draw as much current as it is allowed by the PSE. During startup, a PSE allows current flow of up to 500ma for 100ms, then current limits at 350ma for continuous operation. During the time the PSE is in current limit, the bus voltage drops to the level required to maintain the maximum current. If current exceeds either current limit point, the PSE will shut down and re-enter discovery mode.
Method B: Inrush limiting in the PD
Here, the PD current limit is set below that in the PSE. This means that a PD will limit its own current to a level necessary to maintain the bus voltage at its proper level. During startup, the PD limits current to less than 500ma for 100ms, and limits at 350ma thereafter. The PSE only limits current if there is a fault (short, or mis-wiring) between the PSE and the PD.
In both approaches, I assume the DC/DC converter has its own current limit circuit. This means that if a fault in the load occurs, the DC/DC converter will limit the power to the fault. It may be any type of current limit (self resetting, latch off, or any type the PD designer wants to use).
Observation in operation: I looked at three modes of operation: Startup, A short circuit on in the Ethernet wiring, and finally a short between the switch and the DC/DC converter ( a shorted Bulk Capacitor).
1) Startup: This is when the PSE switch FET turns ON and the output voltage increases from detection/classification to full ON 44 to 57 volts. When the voltage at the PD reaches this regulation, the switch in the PD turns ON.
Method A.
In Method A. the PD switch turns ON fully immediately. Thus, there is a time where the bus voltage will drop as the bulk capacitor in the DC/DC is charged.. To keep the switch in the PD ON, we need to add switched in energy storage at PD - The PD needs to stay alive for a period of time while the input goes away and slowly returns. This storage cannot be there during detection, therefore the cap must be switched in at >30 volts. The size of this capacitor should be large enough to sustain the PD control circuit during this interval, yet small enough to that charging it does not add significantly to the overall startup time. 10uf?
The PSE switch FET is turned ON in a linear fashion, so the turn on dissipation is in the PSE FET in this method.
Also, there needs to be an accurate Under Voltage Lock Out (UVLO) in the DC/DC converter. The DC/DC must stay off until the voltage is completely up, otherwise the system will not start. This is especially true of higher powered PDs.
Lastly, there is a timer required in PSE to determine whether there is a fault, or startup condition. (more on faults later)
Page two of the attached .pdf file shows (starting at the bottom) the input to the PD in blue, and the voltage on the bulk capacitor. the initial "bumps" in the wave form are the voltages seen by the PD during Discovery and classification. At about 40ms the switch on the PSE is closed, and the voltage rises abruptly. Moments later, the switch on the PD closes and the voltage drops, and then gradually increases as the bulk capacitor charges. The top trace, in green, shows the DC/DC converter powering up as soon as the capacitor voltage reaches 44 volts.
Lastly, the middle trace shows the current delivered by the PSE during startup.
Page three shows the power dissipated in the PSE and PD switch FETs.as they are turned on and the bulk capacitor is charged.
Method B.
In this method, the switch in the PSE turns ON quickly, and the PD FET turns ON in a linear mode, limiting the current to a value determined by the PD designer. When the voltage ramps up on the bulk capacitor, the PD FET turns fully ON. In this method, the startup power dissipation is in the PD FET.
The DC/DC converter may be held OFF during the charging of the capacitor, or may be allowed to turn ON. If left to turn ON, the turn on time increases and the dissipation in the series FET increases for that period of time it takes the capacitor to charge.
Page three of the attached .pdf file shows the input to the PD, the bulk capacitor voltage, the PSE current, and the DC/DC converter's output voltage.
In this simulation, the PD enables the DC/DC converter when the capacitor is charged, and so no UVLO circuitry is required. Also, the PD charges the capacitor with a lower current than full load operating current (about 175ma here) to keep power dissipation low in the PD switch.
Page four shows the power loss in the PSE and PD switches during power ON. The PSE switch power dissipation is negligible, and the PD switch loss is less than that seen in Method A. A further decrease in power loss could be accomplished at the expense of a longer capacitor charging time.
NOTE: During simulations, it was observed that Method A needed to start in as short a time to guarantee operation. This limits losses in the event of a fault, but also limits the size of the bulk capacitor used in the PD. Timing of events was critical to operation.
By contrast, in Method B, startup time had no effect the systems ability to start. Whether the time was long to reduce peak power loss, or very short, the system would always start.
2) Shorted wiring: In this operating condition, I am assuming a failure has taken place, which puts a short on the Ethernet cabling.
Method A:
The PSE switch stays ON, delivering 350ma for 100ms, with up to 57 volts across it. ~20 watts until the PSE decides there is a fault and shuts OFF.
Additional circuitry could be added to the PSE which would monitor whether the bus voltage had once been valid for a time, and has become invalid due to a short.
Method B:
If the wiring shorts, the PSE delivers 350ma for 50us or so, then shuts off. The result is the same power dissipation for a much shorter period of time.
3) Shorted bulk capacitor: In this operating condition, I am assuming a short has been placed across the bulk capacitor; between the DC/DC converter and the PD switch FET.
Method A:
Since the PD switch is fully ON, there is a small difference in dissipation in the PD switch FET. In operation, if the PD controller loses its bias power, and shuts OFF the FET. The PD goes to zero power and the PSE enters discovery mode. Otherwise, the PSE will time out and dissipation is the same as in Case 2.
Method B:
The PD detects an increase in current while the voltage on the bus has gone to
zero. The PD then enters a low power mode similar to that of turn ON.
In either Case 2 or Case 3, the PSE will determine that the PD has gone away, and will react accordingly.
Design considerations: When implementing these both circuits in simulations, I noted some of the design considerations, considering PDs ranging in power from ~ 1 watt to the full ~13 watts.
Method A:
a) PSE dissipates power during startup and faults. PSE sizes FET for worst case, not knowing what PD will be used. The time to shut OFF is subject to the size of the bulk capacitor being used in the PD.
b) PD designer needs a startup circuit for the PD control circuitry. The storage in the PD must be long enough to keep the PD alive during startup.
c) PD switch FET has limited power dissipation and can be small.
d) PD designer needs to insure DC/DC converter is OFF until bulk capacitor is charged, otherwise the circuit will not start.
e) The PD designer is limited to the size of the bulk capacitor used in the PD. Too large, and the PSE will time out and the system will not start
f) You can always add inrush limiting to the PD to limit the FET size. In effect, if we specify Method A, then Method B could always be implemented to insure startup operation. The down side is that the PSE is now severely oversized.
Method B:
a) PSE power dissipation is small, so the FET may be kept small for any application.
b) PD designer has freedom to scale his FET size, based on his bulk capacitor size and power level in his own application.
c) Circuit operation voltage and current dependent is independent of circuit time constants.
d) In practice, the inrush current limit in the PD will be set much lower than 350ma. (or the 500ma for 100ms for that matter). By setting the charging current
of the bulk capacitor to say, 100ma, peak power dissipation in the PD switch PET can be significantly reduced (at the expense of a longer capacitor charge time)
I know this is long, but I hate to think of us getting too far behind on some of the issues. Any comments from others?
Brian T. Lynch
Principal Member of Applications Development
Power Supply Control Products
Texas Instruments Incorporated
+ brian_lynch@ti.com
7 Continental Boulevard ( 603 429 6054
Merrimack, NH 03054-4303 2 603 429 8564