LVPECL termination

[Tom Derham]

Normally LVPECL receivers expect termination to be 50 Ohm to Vtt (which is
Vcc - 2V, or 1.3V for normal 3.3v lvpecl).

However, I have seen a couple of evaluation boards (one of which I rather
naively followed for one of my board designs) where both (differential)
outputs are terminated into 50 ohm to ground.
The signals are then used to drive the differential inputs to a DDS chip
(high impedance) requiring essentially lvpecl levels (centred on 1.6v dc
approx).

Can anyone suggest what the effect of this difference in termination will
have on the circuit?
Is it correct / acceptable design practice?

Many thanks

Tom Derham

 

[John Larkin]

Normally LVPECL receivers expect termination to be 50 Ohm to Vtt (which is
Vcc - 2V, or 1.3V for normal 3.3v lvpecl).

However, I have seen a couple of evaluation boards (one of which I rather
naively followed for one of my board designs) where both (differential)
outputs are terminated into 50 ohm to ground.
The signals are then used to drive the differential inputs to a DDS chip
(high impedance) requiring essentially lvpecl levels (centred on 1.6v dc
approx).

Can anyone suggest what the effect of this difference in termination will
have on the circuit?
Is it correct / acceptable design practice?

Many thanks

Tom Derham

Well, the drivers will have to source about 50 ma when they're high,
which does burn up a bit of power but is probably OK... check the
driver spec to be sure. I'd personally go for a termination that
wastes less power and makes the driver work less.

In real life, with Vcc just 3.3, this should work, which is I suppose
the definition of correct/acceptable.

John

 

[Keith R. Williams]

Normally LVPECL receivers expect termination to be 50 Ohm to Vtt (which is
Vcc - 2V, or 1.3V for normal 3.3v lvpecl).

However, I have seen a couple of evaluation boards (one of which I rather
naively followed for one of my board designs) where both (differential)
outputs are terminated into 50 ohm to ground.
The signals are then used to drive the differential inputs to a DDS chip
(high impedance) requiring essentially lvpecl levels (centred on 1.6v dc
approx).

Can anyone suggest what the effect of this difference in termination will
have on the circuit?

I'd think the drivers would be working rather hard.

Is it correct / acceptable design practice?

I wouldn't let this go.

There are a number of options here. I've used a split-terminator with
83ohms to ground and 125ohms to +3.3V quite successfully. In your case
this is simply a value change and the addition of the upper resistor.

 

[Daniel Lang]

Normally LVPECL receivers expect termination to be 50 Ohm to Vtt (which is
Vcc - 2V, or 1.3V for normal 3.3v lvpecl).

However, I have seen a couple of evaluation boards (one of which I rather
naively followed for one of my board designs) where both (differential)
outputs are terminated into 50 ohm to ground.
The signals are then used to drive the differential inputs to a DDS chip
(high impedance) requiring essentially lvpecl levels (centred on 1.6v dc
approx).

Can anyone suggest what the effect of this difference in termination will
have on the circuit?
Is it correct / acceptable design practice?

At 3.3 volts, the high level current will be about 50 mA which is near
the absolute maximum for ECL and wastes power.

I would use a Y termination at the receiver using 50 ohms from each of
the lines to a third 50 ohm resistor to ground.

Daniel Lang

 

[Mac]

Normally LVPECL receivers expect termination to be 50 Ohm to Vtt (which is
Vcc - 2V, or 1.3V for normal 3.3v lvpecl).

However, I have seen a couple of evaluation boards (one of which I rather
naively followed for one of my board designs) where both (differential)
outputs are terminated into 50 ohm to ground.
The signals are then used to drive the differential inputs to a DDS chip
(high impedance) requiring essentially lvpecl levels (centred on 1.6v dc
approx).

Can anyone suggest what the effect of this difference in termination will
have on the circuit?
Is it correct / acceptable design practice?

Many thanks

Tom Derham

I wouldn't think it's acceptable at all. You should check the data sheets
for both the receiver and driver to see if you are violating anything.

A couple of the receivers I've seen have a weak drive on one of the pins,
which would be enough to set the common mode voltatge. In that case, you
can just put 100 Ohms across the receive terminals.

Another option is to use 50 Ohms from each receive terminal to AC ground
(a capacitor). This has the advantage of greatly attenuating any common mode
signal that may be present. But something always has to set the common
mode voltage. Either the driver or receiver, or a pullup/pulldown network.

In your case, what you have to hope for is that the driver will be happy
providing the fairly hefty DC currents you probably need to keep the
receiver in its rated common-mode voltage.

Mac
--

 

[Fred Bloggs]

You saw no such thing- your reverse engineering, aka copying , is in
error...

 

[John Larkin]

A couple of the receivers I've seen have a weak drive on one of the pins,
which would be enough to set the common mode voltatge. In that case, you
can just put 100 Ohms across the receive terminals.

Another option is to use 50 Ohms from each receive terminal to AC ground
(a capacitor). This has the advantage of greatly attenuating any common mode
signal that may be present. But something always has to set the common
mode voltage. Either the driver or receiver, or a pullup/pulldown network.

The logic here is ECL: open emitter outputs. This *must* have a
substantial DC pulldown to work: the driver can only pull up.

50 ma is a tad high, but you'd need at least 20 to work properly here.


John

 

[Mac]

The logic here is ECL: open emitter outputs. This *must* have a
substantial DC pulldown to work: the driver can only pull up.

50 ma is a tad high, but you'd need at least 20 to work properly here.


John

Well, it's LVPECL, not ECL. And a lot of LVPECL stuff is really just
differential. And as far as I know, LVPECL doesn't mean open-emitter, it
just means emitter-coupled. That is, there's no reason the pulldowns
couldn't be on the driver. But the bottom line is that he has to look at
the datasheet for the driver and receiver to see if he is violating
anything.

FWIW, I made a cursory survey of a few LVPECL parts, and the ones I saw
actually can drive 50 Ohms to ground.

Mac
--

 

[John Larkin]

Well, it's LVPECL, not ECL. And a lot of LVPECL stuff is really just
differential. And as far as I know, LVPECL doesn't mean open-emitter, it
just means emitter-coupled. That is, there's no reason the pulldowns
couldn't be on the driver. But the bottom line is that he has to look at
the datasheet for the driver and receiver to see if he is violating
anything.

LV means low voltage. PECL means ECL that is operated with shifted
supplies, ie Vee = ground. If you take any old ECL part and wire its
Vcc and Vee pins to V+ and ground respectively, then you call it PECL.
It's still open emitter. I don't know of any ECL (or 'LVPECL') parts
that have internal pulldowns. LVDS drivers, of course, pull up and
down.

There may be some parts that advertise 'LVPECL output levels' and do
drive up and down, but they're not actually ECL parts. Anybody know of
any?

FWIW, I made a cursory survey of a few LVPECL parts, and the ones I saw
actually can drive 50 Ohms to ground.

Right. 50 mA isn't really a lot for these brutes.

John

 

[Mac]

LV means low voltage. PECL means ECL that is operated with shifted
supplies, ie Vee = ground. If you take any old ECL part and wire its
Vcc and Vee pins to V+ and ground respectively, then you call it PECL.
It's still open emitter. I don't know of any ECL (or 'LVPECL') parts
that have internal pulldowns. LVDS drivers, of course, pull up and
down.

Well, there is a sizeable minority who say the 'P' in PECL is for
"pseudo," not "positive." I'm not one of that minority, mind you.

There may be some parts that advertise 'LVPECL output levels' and do
drive up and down, but they're not actually ECL parts. Anybody know of
any?

Certainly receivers.

The xilinx virtex II fpga can use some kind of "PECL" outputs. I've never
looked into it because I never needed to, but I bet they're not really
PECL. I mean, I assume there is only one die in the package, and it is
130nm CMOS. Can they mix bipolar and CMOS on one die?

Right. 50 mA isn't really a lot for these brutes.

John

Mac
--

 

[John Larkin]

Well, there is a sizeable minority who say the 'P' in PECL is for
"pseudo," not "positive." I'm not one of that minority, mind you.

It used to be 'pseudo', as in the old Motorola books. But some PR type
must have decided that sounded funny, so it's been rehabilitated to
'positive'. I still call it 'pseudo'. I did finally stop calling
capacitors 'condensers', though.

Certainly receivers.

Receivers drive down? I thought receivers, er, received.

John

 

[Mac]

It used to be 'pseudo', as in the old Motorola books. But some PR type
must have decided that sounded funny, so it's been rehabilitated to
'positive'. I still call it 'pseudo'. I did finally stop calling
capacitors 'condensers', though.

Aha! Thanks for clearing that up for me. I have never heard anyone call a
capacitor a condenser. ;-)

Receivers drive down? I thought receivers, er, received.

Well, the only thing I can think of is that I was responding to the "not
actually ECL" part of your question. But I concede that that part of my
reply makes no sense.

I mean, a lot of differential receivers do have weak pulldowns. And they
can detect when the signal is not otherwise driven. When they detect that
condition, they drive their outputs to a logic low. But this isn't what I
was talking about, I don't think.

John

regards,
Mac

 

[John Larkin]

Aha! Thanks for clearing that up for me. I have never heard anyone call a
capacitor a condenser. ;-)

Ah, a youngster you are (in Yoda-speak.) But there are still a few
cars and motorcycles with 'points', and they still have 'condensers'.

Somewhere around here I have a really handy CPS-to-Hertz nomograph...
Yes! I'll post it to a.b.s.e.

John