Thread Links | Date Links | ||||
---|---|---|---|---|---|
Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
Roger,
I'm probably over-reacting here....
As you know, there are many ways to design passive filters, PI, T, etc...
For lowpass we are talking about series L's and shunt C's.
You can choose large L's and small C's, or visa versa.
The real choice will be driven by economics,
where the L's tend to be rather small (nH or uH), and the caps tend to be large (nF or uF)
in the frequency range we are talking about.
For the purposes on the diode discovery using the transformers that has been shown, the output impedance
of the power supply filter is in series with the loop.
So if that output filter has a relatively low impedance below 100 KHz, the diode detection will work.
I maintain that when you design a practical lowpass filter to tame the power supply so that it does not corrupt the data,
it will also work for the diode discovery, since there will be a shunt cap and therefore low output impedance.
Noise below 10 KHz on the output of a DC power supply, is controlled by feedback and output capacitance, not brute force filters.
Noise from 10 KHz to 1 MHz is controlled mostly by output capacitance, along with some series inductance.
This is driven by economics.
It is desirable and quite natural to have a low output impedance from a constant voltage DC power supply.
Yes, you can design a filter so that the diode discovery using transformers does not work, but it will probably be costly.
- Rick
PS: I always agree with you that we need to do lab work to back things up.
-----Original Message-----
From: R karam [SMTP:rkaram@xxxxxxxxx]
Sent: Friday, September 22, 2000 9:40 AM
To: Brooks, Rick [SC5:321:EXCH]; Robert H Leonowich
Cc: 'IEEE 802.3 'DTE Power''
Subject: Flying Responses
Hi
Just wondering if the comments about how the supply filter would affect
the AC pulses path has been analyzed in the Lab, and the same for the resistor
scheme, early on in this game we had a lot of flying assumptions, gut feels,
and
affection for positions that did not pan out with hard work on the bench.
the assumption of a clean supply is a good one, but by the time the power
travels
through connectors in a real system, it may not be as clean as we think, so I
would
say making sure that heavy duty filtering can be placed in series with power
without
disabling the detection scheme is not a bad idea, and Please let's not only
spice things
let's lab them.
2c,
roger
At 07:30 AM 9/22/00 -0700, Rick Brooks wrote:
>
> Bob,
> here are my comments on your comments:
>
> 3.) Logic side power plane residence of detection implies EMI can leak from
> noisy logic to wire response: The exact same thing is true of the 10/100/1000
> data paths. EMI problems will be dominated by the switching power supply rhl:
> It seems the common-mode EMI blocking of the data path's transformers should
> be different than that from the signature transformer's?
> rb: the signature transformers pass 100 KHz signals, so the common mode
> rejection is good, at high clock frequencies
> routing or filtering might be required. This could be incorporated into the
> drivers.
>
> 4.) Approximately 1MHz bandwidth signature pulse signaling precludes
> aggressive filtering to limit noise emissions response: If you look at page
> 27 of my presentation, the common mode discovery pulse energy is about 40 dB
> down at 1 MHz, the 3 dB point is more like 30 KHz. rhl: To clarify, I meant
> without a bandwidth constraint in series with the power supply, the user is
> free to filter at any corner frequency to eliminate noise from the supply
> when/if that becomes an issue down the road. If the AC pathway must be kept
> clear, that is one more obstacle to aggressively suppressing power supply
> noise.
> rb: My view is that you can filter the DC power supply where ever you want.
> The diode detection
> counts on having an output cap on the power supply to provide a low impedance
> path in both directions through the loop.
>
> 5.) Conveyance of current use from supply to upper levels (to facilitate
> network power management) requires crossing the isolation barrier and
> additional monitoring circuits in the supply response: This is not tied to
> the discovery method. rhl: However, there seems to be an additional cost
> associated with providing this function in the diode scheme (as presented)
> since the increased cost of the power module with and without this function
> was not given.
> rb: As you have framed the scheme (PHY based detection, PHY or logic
> managed), yes you need these (at least some) control signals.
>
> That does not have to do with the detection method however, the method can be
> done in the power supply as well.
> But are you suggesting putting the mgmt in your detector chip or the power
> supply?
> How do you communicate with your power supply and detector?
>
> 7.) Parallel devices may get destroyed at application of higher power
> response: Can you be more specific about which parallel devices you are
> talking about? rhl: This attempts to address the devices outlined in Bob
> Bell's liaison input to us from: TIA TR41.4 Subcommittee on VoIP Gateways
> and Infrastructures
>
> "The ISDN system might legitimately have multiple S-Bus devices on the same
> 4-pair cable which terminates using "RJ45"-type connectors. If that cable is
> reused for an 802.3 network, the endpoint might very well respond to the
> detection scheme within the detection parameters but when power is applied,
> parallel connected surge protectors left over from the ISDN usage might
> overheat and cause a fire if subjected to the nominal voltage and if the
> current is not limited correctly."
>
> It also can pertain to test equipment that does not appreciably conduct at
> low voltage (<2V) but does at higher voltage (>6V). A higher signature
> voltage allows these devices to be detected before having 15W of power
> applied.
> rb: I doubt if there will be a positive response to the diode detection
> scheme, if true, this point is moot.
>
> 11.) 1uF >54V signature cap in PE is big and expensive response: The cost
> needs to be considered. Also, the cap value can probably be lowered. Even so
> a 1206 part is not that big. rhl: The DTE would appear to need 2 of these
> cap/signature modules for power on signal/spare pair with the diode scheme.
> This is not true on the resistive scheme.
> rb: I agree with you here.
> We should probably just pick one pair and be done with it anyway, this would
> mean only one identity networks,
> and less diodes required.
>
>
> thanks,
> - Rick
>
>
>
> -----Original Message----- From: Robert H Leonowich
> [SMTP:leonowich@xxxxxxxxxx] Sent: Thursday, September 21, 2000 3:43 PM
> To: Brooks, Rick [SC5:321:EXCH] Cc: 'IEEE 802.3 'DTE Power''
> Subject: Re: response to Bob Leonowich's presentation, page 17, Sept
> 2000
>
> Hi Rick,
>
> Here are some response comments:
>
> 3.) Logic side power plane residence of detection implies EMI can leak from
> noisy logic to wire response: The exact same thing is true of the 10/100/1000
> data paths. EMI problems will be dominated by the switching power supply rhl:
> It seems the common-mode EMI blocking of the data path's transformers should
> be different than that from the signature transformer's?
>
> 4.) Approximately 1MHz bandwidth signature pulse signaling precludes
> aggressive filtering to limit noise emissions response: If you look at page
> 27 of my presentation, the common mode discovery pulse energy is about 40 dB
> down at 1 MHz, the 3 dB point is more like 30 KHz. rhl: To clarify, I meant
> without a bandwidth constraint in series with the power supply, the user is
> free to filter at any corner frequency to eliminate noise from the supply
> when/if that becomes an issue down the road. If the AC pathway must be kept
> clear, that is one more obstacle to aggressively suppressing power supply
> noise.
>
> 5.) Conveyance of current use from supply to upper levels (to facilitate
> network power management) requires crossing the isolation barrier and
> additional monitoring circuits in the supply response: This is not tied to
> the discovery method. rhl: However, there seems to be an additional cost
> associated with providing this function in the diode scheme (as presented)
> since the increased cost of the power module with and without this function
> was not given.
>
>
>
> 7.) Parallel devices may get destroyed at application of higher power
> response: Can you be more specific about which parallel devices you are
> talking about? rhl: This attempts to address the devices outlined in Bob
> Bell's liaison input to us from: TIA TR41.4 Subcommittee on VoIP Gateways
> and Infrastructures
>
> "The ISDN system might legitimately have multiple S-Bus devices on the same
> 4-pair cable which terminates using "RJ45"-type connectors. If that cable is
> reused for an 802.3 network, the endpoint might very well respond to the
> detection scheme within the detection parameters but when power is applied,
> parallel connected surge protectors left over from the ISDN usage might
> overheat and cause a fire if subjected to the nominal voltage and if the
> current is not limited correctly."
>
> It also can pertain to test equipment that does not appreciably conduct at
> low voltage (<2V) but does at higher voltage (>6V). A higher signature
> voltage allows these devices to be detected before having 15W of power
> applied.
>
> 11.) 1uF >54V signature cap in PE is big and expensive response: The cost
> needs to be considered. Also, the cap value can probably be lowered. Even so
> a 1206 part is not that big. rhl: The DTE would appear to need 2 of these
> cap/signature modules for power on signal/spare pair with the diode scheme.
> This is not true on the resistive scheme.
>
> Bob
>
>
>
> > Rick Brooks wrote: > > Bob, > I'm sorry that I was not at the meeting to
> answer these questions. > Here are my responses to the item numbers on your
> presentation, page 17 > > 1.) Transformer doesn't necessarily provide AC/ESD
> impulse isolation so this > needs to be > tested > > response: > ESD was
> tested informally, but not reported. > Recall that the logic side of the
> transformer has a 100 ohm and a 3300 pf cap > to ground. > This shunts the
> ESD energy that gets through the transformer. > This is much easier than on
> the data lines, since the pulse bandwidth is so > much lower. > > 2.)
> Strategy implies PHY redesign and consequential re-qualification on new >
> generation > devices with higher pincount to support the needed I/O > >
> response: > My method can be implemented in the PHY, or in the power supply,
> whichever is > more cost effective. > My prototype shows a PHY
> configuration. > > 3.) Logic side power plane residence of detection implies
> EMI can leak from > noisy logic to wire > > response: > The exact same
> thing is true of the 10/100/1000 data paths. > EMI problems will be dominated
> by the switching power supply > > 4.) Approximately 1MHz bandwidth signature
> pulse signaling precludes > aggressive filtering to > limit noise emissions
> > > response: > If you look at page 27 of my presentation, the common mode
> discovery pulse > energy is about 40 dB down > at 1 MHz, the 3 dB point is
> more like 30 KHz. > > 5.) Conveyance of current use from supply to upper
> levels (to facilitate > network power > management) requires crossing the
> isolation barrier and additional monitoring > circuits in > the supply > >
> response: > This is not tied to the discovery method. > > 6.) Longitudinal
> voltage characterization in terms of <1 volt sensitivity has > not been done
> > > response: > What exact tests are you asking for? > > 7.) Parallel
> devices may get destroyed at application of higher power > > response: >
> Can you be more specific about which parallel devices you are talking about?
> > > 8.) Use of constant signaling and reliance on transformer saturation to
> detect > undercurrent is > very coarse. >10mA signature pulses imply a >0.5W
> maintenance consumption > > response: > The prototype we built uses the
> power supply to detect a minimum load. It > could use transformer saturation,
> if > that is practical, I'm hearing that the magnetics vendors are against
> it. > Again, this is a separate issue from the discovery method. > Also,
> what does the amplitude of the signature pulse have to do with power? > When
> the DTE power is on, and the undercurrent value matters, and the > discovery
> pulses are off. > > 9.) 25 pair bundle interference with asynchronous units?
> > > response: > This is on my list of things to test. > > 10.) Pulse
> shaping implies expensive immediate implementation > > response: > This is
> true if you use my discrete approach, but why would anybody do that? > An
> integrated IC solution would use multiple current sources, as the PHY's do >
> now. > > 11.) 1uF >54V signature cap in PE is big and expensive > >
> response: > The cost needs to be considered. Also, the cap value can probably
> be lowered. > Even so a 1206 part is not that big. > > 12.) Extra two diode
> drops Required in PE lowers maximum power delivered to PE > load > >
> response: > We could use a FET switch instead. All discovery methods, except
> the large > cap, face this problem of removing > the effect of the 100 to
> 500 uF cap on the input of the power supply on > discovery. > > any
> comments? > > thanks, > - Rick