RJ45 with integrated magnetics - different configurations

[peter-h]

I posted it here since the other forum sections are not as responsive on complicated matters

The "classic" config, of the "classic chinese HR911105A RJ45", is this



The above is also heavily counterfeited. The Hanrun part is 2x the price of the fakes but much cheaper than say Wurth parts.

It works but has a weakness: if somebody plugs in a cable which is fed by a cheap chinese POE injector (one which simply sticks 48V across the two wires - I forget which two - without any POE negotiation) that melts the 75R resistors instantly.

That leads to this one e.g. Hanrun HR913322A:



which solves that problem, but costs quite a bit more.

Does anyone bother with the capacitor version?

[woofy]

I purchase thousands of magjacks from CONNFLY and Würth and they are both of the 1st type, with the single capacitor.
So far there have been zero reports of damaged magjacks, so not sure how big an issue it really is in practice.

[Nominal Animal]

Typically these "passive injectors" put the positive voltage on J4 and J5, and the negative/ground/return on J7 and J8.
For example, many MikroTik routers can be powered using these, as they are similar to MikroTik's own RBGPOE.

Whenever this kind of "passive injector" is described, it is usually mentioned that this is not proper PoE, i.e. IEEE 802.3af-2003 (PoE), IEEE 802.3at-2009 (PoE+), or IEEE 802.3bt-2018 (PoE++ or 4PPoE), and not compatible with those.  Even in the hobbyist/Arduino environments, it is "known" that using these "passive injectors" can damage standard equipment.  (That is, anyone who asks in a forum or mailing list, will be told it may damage their equipment.)

I would only bother with the separate capacitors version, if I wanted my device to be compatible with these cheap "passive injectors", even though not able to be powered via one of those.

[peter-h]

Interesting - thanks.

I could have also used a jack with magnetics but no resistors, and then adding all these externally costs nothing. But it's too late now; I don't want to change the PCB. I used this RJ45 because the ST 32F4 dev board had it, and that board was used as a starting point.

[thm_w]

Using one similar to the first type, a few failures where you can measure the resistance between pins and you'll get some value other than 150R which means its blown yeah. But very rare, 1 in 500 maybe. More common to see ESD/lightning blow outs. I suppose those could have also caused the issue.

The annoying thing is there are at least 4 or 5 different Magjack pinouts. Some are even the same internal config, just external pins have been swapped around.

[nctnico]

I think the real question here is: how bad is it when the 75 Ohm resistors blow up? If that doesn't affect the wiring inside the magjack in any way, then I'd assume there is not a big (showstopping) problem.

[thm_w]

I think the real question here is: how bad is it when the 75 Ohm resistors blow up? If that doesn't affect the wiring inside the magjack in any way, then I'd assume there is not a big (showstopping) problem.

Good point, just tried applying power across Blue pairs to Brown pairs, blew the resistors inside. Port still works, interesting.
So I guess in my case it was always lightning failures and not passive POE injection.

Although the loss of resistor might make it more susceptible to other failures.

[nctnico]

PoE is not only injected on the blue & brown pairs, but also on orange and green (the signal pairs) and I've seen that causing problems with blown ports as a result as well. The best way to avoid problems is to used a rugged phy, series beads / resistors and TVS diodes. Many ethernet phys only allow 6V, maybe close to 7V on their pins so using the right TVS diodes is very important.

[thm_w]

OK, tried orange and green which is called "802.3af Mode A" here: https://pinoutguide.com/Net/poe_pinout.shtml
Port still working.

[Whales]

This is fascinating to read about.  Passive & non-standard PoE is more common than people imagine, thanks to IP cam installers (and early ubiquity products, but they're 802.3af compliant now).

If those 75 ohm resistors blow then the termination impedance of the ethernet cables ends up wrong.  In MANY cases things will still work, but depending on cable lengths & the devices being used you might not be able to get a reliable link (or your devices might downgrade to a slower link speed).

Next time I have a device with weird ethernet behaviours: I'm going to buzz out the resistance between pins

[ajb]

A friend of mine had an HDBaseT midspan injector that fried the upstream, non-powered device when the upstream, non-powered port was connected after the powered device—twice.  A little reverse engineering revealed that Microsemi had followed their own application schematic to the letter in this capacitively isolated midspan injector. Turns out that 57V through 100nF will wallop a PHY even with a magjack and bob smith termination in the way.

The magjack, including termination resistors, seemed to be no worse for the wear, but the PHY ended up with a low-ish resistance connection between the rx and tx pins.

[radar_macgyver]

and early ubiquity products, but they're 802.3af compliant now).

That's true of the Unifi series, but their EdgeMax devices (meant for small ISPs) use passive PoE at 24V, mainly for their radio products. EdgeMax PoE switches allow you to manually turn on 24V injection on each port.

[T3sl4co1l]

Yep, never underestimate the impulse of a mechanical connection discharging a capacitor.  The risetime is very fast (subnanosecond) and the peak current can be, well probably just amperes in this case, on up to 100s or 1000s of A for electrolytics and, say, ones' hapless screwdriver (or breadboard, or..).

Note that pins in general will mate in some random order (well, maybe it's relatively consistent for a given pair of mates, but in general, you can't know which), so you're essentially guaranteed to get some full-magnitude unbalanced surge across one or both pairs.

Note that an unbalanced discharge does flow through the common mode chokes, so the risetime will be attenuated some; but within a few hundred ns, the core will saturate, and the bulk of the pulse will continue onward to the PHY.  Which will in turn saturate in some ~µs, but in that time, the damage has been done to the PHY.

So yeah, ESD diodes are a must with PoE!

Tim

[peter-h]

For actual intended POE, the RJ45 is a different P/N anyway. It contains diodes and other stuff, IIRC.

I am not implementing POE because it increases the cost a fair bit. My product has two separately isolated-powered sections so I would need two DC-DC converters. Or one driving the main DC input. Then EMC certification stuff to be done all over again. It also BTW needs to be done so that if you have POE and the external power unit (or DC IN) is/are also connected, nothing blows up.

I think POE is big with "remote" devices (wifi APs and such) but not so much industrially which is what I am doing. POE switches can work out very expensive.

And any proper POE switch (e.g. Netgear, Cisco, etc) will not blow anything up.

I would expect blown-up 75R terminators to badly compromise operation, other than with 30cm cables I know USB is different but it totally breaks if you strip the shielding from just 3cm of the cable (as if e.g. wiring a toggle switch so you can use a USB connection for data or for just external DC power).

[mon2]

Sharing some good contacts for RJ45 with magnetics - each builds for major brands under private labels.

Jason Cai   |   Sales Engineer
Jason <jason[at]link-pp.com>
LINK-PP INT'L TECHNOLOGY CO., LIMITED
Tel:  +86-752-3322917       |    Mobile: +86-13434796102
Wechat :  jasoncai1221    |       Skype: link-pp6
Add: Bldg A-B No.54, South of Jinhu Road,Chengjiang Town, Huizhou City, 516229, China
RJ45 Connector | Lan Transformer | Optical Transceivers  | SFP+ cage & Conn

sherry
Company:Trxcom INT'L Technology Co. Limited.
Tel:0086-755-81755632
Fax:0086-755-81752963
Cel:0086-135-5491-8626
Email: sales1[at]trxcom.com
Skype: bettytrxcom
Web: www.trxcom.com
Trxcom is the official vendor for Raspberry Pi 4 PoE project
Reference Link: https://www.raspberrypi.org/products/raspberry-pi-4-model-b/

[westfw]

If those 75 ohm resistors blow then the termination impedance of the ethernet cables ends up wrong.

huh?  If they're using passive PoE, then by definition those pairs already had horrible termination impedance, and it didn't matter because nothing was using them for data anyway, and the devices were by definition not using them for any kind of PoE, either (or they'd need a different magjack.)

I suppose there is potentially a problem if you move the devices with the blown resistors to a proper network without the passive PoE; a switch might get confused.  But the device is still not using those pairs, and I'd expect the normal switch behavior would be to ignore them as well (after all, they ought to ignore a less-than-4pair cable that was only doing 10baseT, right?  Or a full cat6 cable connected to a 10Mbit only device that doesn't implement those pairs at all (should be electrically the same as the Magjack with blown resistors, right?)

[cdev]

I recently got my first POE switch which I plan to use for IoT ish stuff, Its got some issues so don't plan to use it for my main LAN, I have yet to thoroughly check it out, are there any failure modes that crop up often? One nice thing it can do is NTP, it appears! (one can even connect a GPS to it) It can do hardware time distribution which I have wanted for a long time. And of course it can power VOIP phones.. and do QOS

[Whales]

and do QOS

N.B. have a read up on bufferbloat.  QoS is basically useless if the packets are intended to be sent over the internet and your internet link is externally saturated. 

Learning this magic made all the difference in the world for me, and explained lots of weird oddities I've always blamed on my home router or ISP over the years. 

[peter-h]

one can even connect a GPS to it

Which model is it?

POE is indeed very good for VOIP phones, and WIFI APs and such like. But all those are "proper POE" devices which don't need the cheap chinese POE injectors which are the real problem. Also any POE switch will be doing proper POE so no power comes out unless negotiated.

[peter-h]

For amusement, real and fake.
The funny thing is that the real Hanrun, being chinese, is probably a counterfeit of another one

The fake one is marked F. The real one also has "Hanrun" inside it at the far end (visible in photo).

The fake one is 5.5g and the real one is 6.0g.

Both seem to work equally.

[Emo]

Hi Peter,

The GPS is connected to the USB port. Mikrotik is then able to use it as NTP

Eric

[peter-h]

Ah it must be a special GPS, with USB. Which Mikrotik box are you using and which antenna? I can't find them on their website.

[nctnico]

Ah it must be a special GPS, with USB. Which Mikrotik box are you using and which antenna? I can't find them on their website.

Likely all you need is a USB to serial adapter between the GPS receiver and Mikrotik.

[Doctorandus_P]

It's not really related because the photograph below does not have integrated magnetics, but it does show there is some weird stuff out there.
Presumably some manufacturer read somewhere the wires should be twisted.

The picture is not mine. Just found it on the 'net some years ago.

[Nominal Animal]

Presumably some manufacturer read somewhere the wires should be twisted.

Considering the numerous different pinouts for RJ45 connectors, I think it is more likely that the manufacturer created a connector without magnetics using the footprint and pinout of an existing RJ45 connector that does have the magnetics.

[T3sl4co1l]

Why they didn't also swap 3 and 6, who knows.

Tim

[peter-h]

More like somebody had 10 million PCBs made with the wrong pinout

[ajb]

That weirdo crossed over connector doesn't make sense for ethernet.  Any relatively modern PHY won't care about polarity within the pairs, and the 4/5 and 7/8 pairs aren't used in 10/100, so why bother crossing them at all?  Unless it's gigE, which certainly wouldn't care about pair polarity.  Maybe some sort of non-standard PoE, and it was easier to manufacturer the connector symmetrically and swap both outside pairs?

There are plenty of other uses for 8P8C connectors, like RS485, RS232, etc, so I would guess it's for something like that.  Maybe a console connection where someone forgot to swap the DTE/DCE ends between two designs?

[ajb]

For actual intended POE, the RJ45 is a different P/N anyway. It contains diodes and other stuff, IIRC.

The only necessary difference between a magjack that can be used for PoE and one that can't (aside from the termination difference that you've noted) is that for PoE you need access to the line-side center taps on the transformer (and the unused pairs if it's not a gigE part).  I suppose there's no fundamental reason you couldn't include the rectifiers in the magjack as well, but I don't recall ever seeing that.  Unless you mean TVS diodes, which, same situation--possible, but I don't think it's common.

If you had a product with both PoE and non-PoE variants, the easiest solution is probable to use the same PoE-compatible magjack on both and make all of the other parts DNI.  Or use a PoE module that plugs in to the standard board.

[Doctorandus_P]

The only necessary difference between a magjack that can be used for PoE and one that can't  is that for PoE you need access to the line-side center taps on the transformer.

Are you sure about that?
I have never used PoE myself, but I'd guess that the transformer windings must be able to handle the current, and do so without saturating the magnetics.
Core saturation is maybe not a problem, depending on how the transformer is wired, but it must be able to pass the current, so wires must be thick enough.

[peter-h]

Very true about saturation of the ferrite.

In most apps, the current won't be much because at 48V you can power quite a lot of stuff. A typical product might be a VOIP phone or a WIFI AP and they will be drawing a few W max i.e. below 100mA.

[westfw]

VOIP phone or a WIFI AP and they will be drawing a few W max i.e. below 100mA.

That’s very optimistic, since iirc, voip phones of increasing complexity were one of the things that kept the standard pushing up total power delivery up to 60W and more…
The “default” is 15w…

[peter-h]

That 15W has to be dissipated into heat however... I have a SNOM D375 on my desk (just an example; I would not recommend this POS to anybody) which can't be drawing more than a few watts. 60W would "melt" a VOIP phone

[Whales]

Now you have given me mental horrors about fan-forced VOIP phones.


"The outlet air grill has an aiming mechanism just like car AC.  You can point it away from you if it makes you uncomfortable"
"We keep trays under every phone in case the fans seize up and they melt again."
"Most of their processing power is used emulating java 7 inside an embedded web browser"
"No it still can't properly do early media, people have to say hello twice"

[spostma]

a bit late reply, but possibly useful.

As you cannot know in advance whether the uplink switch to be used with passive Poe is surge protected or not,
it could be useful to add a dc power ferrite choke in series with each DC power line insde the passive PoE injector.

After all, the PoE injector should be low-ohmic on DC signals, not on AC.

If you fear EM emission problemss, then a small 22R + 1nF snubber would be enough to suppress power supply emissions IMHO.

[hedley]

My company spent months trying to resolve the following issue - 10/100 ethernet worked fine on our new board . When we connected to POE or POE + ,  the link became unreliable after 50 meters of cat5e cable between edge device and POE switch . On board we have a WE 749012011 POE lan transformer . After 6 months of head scratching , pcb layout changes , specking to Wurth locally and in Germany ,  we had a response from an WE- FAE in Japan - " just flip the transformer so the common mode chokes are on the PHY side , and not the lan side,  as shown in many app notes and datasheets - Problem solved and we have 1000's of units in the field BUT I have no idea why it worked.

[peter-h]

You must have been using an RJ45 without integrated magnetics.

[Nominal Animal]

we had a response from an WE- FAE in Japan - " just flip the transformer so the common mode chokes are on the PHY side , and not the lan side,  as shown in many app notes and datasheets

Würth's own datasheet (from e.g. mouser.fi) does show the chokes on the PHY side, though.

You must have been using an RJ45 without integrated magnetics.

Sure, but it does illustrate why integrated magnetics are a good thing: one less thing to worry about.

There are, in fact, RJ45 female connectors with integrated magnetics and POE taps, for example Abracon ARJM11xx-114 for 100Base-T and POE+.  Mouser has these in stock for under 3.5€ apiece, and they cost less than the Würth transformer above (at Mouser at least); both having similar current capability (600-720 mA).  I have no idea about the quality or reliability, though, just pointing them out.

[peter-h]

Wurth stuff (German i.e. chinese) is extremely expensive. I doubt anybody uses it in production. Chinese e.g. Hanrun are 1/4 of the price, and then you get counterfeit Hanrun which are half the Hanrun price (but work just fine).

The Abracon ARJM11A1-009-NN-EW2 (£1.61 at Mouser, 1k+) is about 2x the price of a Hanrun HR911105 one which has an identical circuit; that's perhaps the most common non-POE layout around.

These will get blown up by a cheap chinese POE power injector.

The Abracon ARJM11D7-114-AB-EW2 (£1.74 at Mouser, 1k+) should not get blown up.

Personally I would not use Wurth or Abracon due to the price. Both are chinese and rebadged while lifting the buy price about 4x

[Ice-Tea]

Wuerth is expensive but especially so if odering through Farnell or other distris. If you have a decent project and contact them they discount deeply. Obviously not as deep as the chinese but, well, you get what you pay for.

[peter-h]

you get what you pay for.

What would be the benefit of a Wurth versus Hanrun?

I mean, nobody in Germany will be making these. Wurth must be badged chinese product. The only Q is whether it is Hanrun or one of the lesser known ones e.g. Hulyn which are half of the Hanrun price.

I have built boards with Hanrun and Hulyn and all seem to work exactly the same.

[nctnico]

My experience with ethernet connectors is that they may develop contact problems over time. So I like to use connectors from a good brand with a decent layer of gold plating. 99% of problems in a network are due to contact issues.

And for sure Wurth stuff is made in China like anything else but I'd assume there is at least some quality control. In my designs I use a mix of various brands (typically TE, Molex or Wurth). Depending on what is available and pricing.

[Ice-Tea]

you get what you pay for.

What would be the benefit of a Wurth versus Hanrun?

*shrugs*

Consistency of quality? Hi-pot testing that was actually done *and* checked? Gold plating that's actually as thick as what it says on the datasheet? Plastic that's not faul yellow after a year? Local stock? Support? A correct invoice? Someone that understands your language?

All of the above?

The candy the reps something bring you? 

[nctnico]

Haribo macht Vati froh 

[Nominal Animal]

Personally I would not use Wurth or Abracon due to the price.

Me hobbyist caveman.

The Abracon was the cheapest RJ45 with both magnetics and POE+ support that Mouser sells.  I can't see any way to find them at e.g. LCSC, other than checking each of the 2500 or so connectors they have, or by first looking up a manufacturer and part, and then checking if LCSC sells them.  Looking at the 40 or so HanRun RJ45 they do have in stock, HY931147C seems to be the only POE compatible one.  It even includes the full bridge rectifier diodes, although it is probably limited to max. 350 mA (IEEE802.3af).

[JPortici]

you get what you pay for.

What would be the benefit of a Wurth versus Hanrun?

*shrugs*

Consistency of quality? Hi-pot testing that was actually done *and* checked? Gold plating that's actually as thick as what it says on the datasheet? Plastic that's not faul yellow after a year? Local stock? Support? A correct invoice? Someone that understands your language?

All of the above?

The candy the reps something bring you? 

keg of beer! speck!
Though that's the mechanics' Rep.
Our Rep brings us nothing but nagging

[peter-h]

Of course; for low volume I use the usual people like Mouser. But for production they cost too much.

[Siwastaja]

Wurth stuff (German i.e. chinese) is extremely expensive. I doubt anybody uses it in production.

I have used WE stuff small scale because the distributor markup for their parts is generally 3x but WE is willing to sell small quantities directly unlike most other manufacturers, so you easily get the price down by 50-70% by just asking a quote.

For example, power inductors I look at Mouser or Digikey, a Coilcraft part costs 3€ and WE part 6€ @ 1000pcs, but when I do a few clicks on WE website, I get free samples delivered plus a quote for 2.50€ @ 1000pcs, and can also buy just 100.

So clearly the stuff is not expensive, it's just more expensive at distributors.

[Siwastaja]

The candy the reps something bring you? 

This, whenever the rep calls me about the new catalogs, I make sure to remind about the most important thing, because one time he forgot the gummy bears. Next time I was compensated by maybe 20-30 of those small bags.

Unironically, this candy thing alone is a decent reason to design in a WE part. We do design choices for much more arbitrary and meaningless reasons after all.

[langwadt]

Very true about saturation of the ferrite.

afaict the current should mostly cancel except for any difference in resistance of the wires in a pair

[thm_w]

You must have been using an RJ45 without integrated magnetics.

Sure, but it does illustrate why integrated magnetics are a good thing: one less thing to worry about.

I think the point being that some of us like to stay on topic 

[Nominal Animal]

Okay.  Me stray.  Sorry.

Back on topic: Am I the only one who finds it odd that some of the integrated magnetics RJ45s have common mode chokes on the cable side (before the isolation), whereas others have them on the PHY side (after the isolation)?

In particular, of the aforementioned Abracon ARJM11 series, -009- (100Base-T) and -502- (1000Base-T) have them on the cable side, but -104- (100Base-T POE+), -114- (100Base-T POE), -805- (2.5GBase-T), -809- (2.5GBase-T), and -811- (5GBase-T) have them on the PHY side.

Why would you have a common mode choke on the cable side anyway, prior to the isolation, instead of on the PHY side?

[nctnico]

I've seen LAN transformers with common mode at the primary (PHY) side and some at the secondary side. On the secondary side I guess you have the advantage that whatever RF leaks from the ground / power connections from the primary into the secondary side, gets filtered.

[T3sl4co1l]

Instinctively to me, media side is better because if you have RF currents coming in, they're choked away from ground, whereas without, that flows right into whatever termination you have (HoJo resistor + ESD cap).  You're relying more on the CMRR of the transformer, something the primary side CMC can't address.

It's less common for PoE, I think because CMC on pairs would trivially saturate; it's only meaningful for pair-pairs at a time, or the whole cable.  So a more complex winding is needed.  Maybe this isn't a hard limit after all -- good working examples of PoE apparently exist* -- but maybe it's harder to do than just tossing a power tap at the line and assuming everything else is normal.

*Mind, I haven't done anything with PoE yet, so my experience there is limited.

Also, especially with respect to GBASE's, there are full balanced vs. push-pull drivers (the latter needs VDDA to center tap), generally GBASE being the former type, so the transformer can be simpler, permitting higher bitrates.

Tim