On Monday, May 17, 2021 at 2:00:42 PM UTC+10, jla...@highlandsniptechnology.com wrote:
> On Mon, 17 May 2021 13:13:46 +1000, Chris Jones <lugn...@spam.yahoo.com> wrote:
> >On 17/05/2021 08:45, jla...@highlandsniptechnology.com wrote:
> >> On Sun, 16 May 2021 22:28:29 +0200, Gerhard Hoffmann <dk...@arcor.de> wrote:
> >>> Am 16.05.21 um 20:22 schrieb jla...@highlandsniptechnology.com:
> >>>> On Sun, 16 May 2021 20:04:58 +0200, Gerhard Hoffmann <dk...@arcor.de>
<snip>
> >> I bailed on the shorted transmission lines to make the short pulse.
> >> They would have been 16 ohms, and were a nightmare to lay out.
> >
> >You could make 3x shorted 50 Ohm lines and connect them in parallel. You
> >might be able to put them on different layers to save area.
>
> We wanted to do a coplanar waveguide, because it would be easy to
> hack, to tune the length of the short. But it was too big.
>
> Vias are nasty at these speeds, so all the fast stuff is on the top layer.
Microstrip is dispersive, which can matter at those speed, so you should realise the transmission line as buried stripline and used just one via to get it out of the board - which is an inevitable discontinuity anyway.
--
Bill Sloman, Sydney
Reply by ●May 17, 20212021-05-17
On Mon, 17 May 2021 13:13:46 +1000, Chris Jones
<lugnut808@spam.yahoo.com> wrote:
>On 17/05/2021 08:45, jlarkin@highlandsniptechnology.com wrote:
>> On Sun, 16 May 2021 22:28:29 +0200, Gerhard Hoffmann <dk4xp@arcor.de>
>> wrote:
>>
>>> Am 16.05.21 um 20:22 schrieb jlarkin@highlandsniptechnology.com:
>>>> On Sun, 16 May 2021 20:04:58 +0200, Gerhard Hoffmann <dk4xp@arcor.de>
>>>
>>>>
>>>> It's astounding that you can get a 10 gbps SFP module for $20 from
>>>> Amazon. I'd quote about $2000.
>>>>
>>>> We use them for testing fast o/e converters.
>>>
>>>
>>> SFP is much easier than XFP. Esp. XFP @ 1550 nm.
>>> SFP could use VCXLs. That means 5 mA against 50 mA
>>> modulation current for DFB, and simple resistors
>>> vs. real bias tees in XFP.
>>>
>>> They could not grow the vertical cavities for the VCXLs
>>> for large wavelengths. Maybe it is better now.
>>>
>>> The AC coupling is a great help, otherwise the
>>> modulator would have to provide the DC bias, too,
>>> with much larger & slower transistors, and constantly
>>> changing the operating point for power and extinction ratio.
>>>
>>> Infineon sold "us", i.e. Infineon Fiber Optics to Finisar,
>>> I spent some weeks with 3 other colleges in San Jose
>>> for tech transfer. Me, as a free lancer, not even a real
>>> IFX member. :-)
>>>
>>> At that time, the real cost was about your $2K for XFP. We
>>> dreamt of $300 or so. But if you have a _pilot_ production line
>>> with 300 people...
>>>
>>> Gerhard
>>>
>>
>> Even longwave SFPs are cheap now. You can also get WDM ones that
>> transmit in both directions over a single fiber.
>>
>> A genuine Cisco 850nm/10g lists for something like $1200 now.
>>
>>
>> Here's my layout so far.
>>
>> https://www.dropbox.com/s/2pmf16ng6wpzgo1/Z498A-7.jpg?raw=1
>>
>> I bailed on the shorted transmission lines to make the short pulse.
>> They would have been 16 ohms, and were a nightmare to lay out.
>
>You could make 3x shorted 50 Ohm lines and connect them in parallel. You
>might be able to put them on different layers to save area.
We wanted to do a coplanar waveguide, because it would be easy to
hack, to tune the length of the short. But it was too big.
Vias are nasty at these speeds, so all the fast stuff is on the top
layer.
--
John Larkin Highland Technology, Inc
The best designs are necessarily accidental.
Reply by Chris Jones●May 17, 20212021-05-17
On 17/05/2021 08:45, jlarkin@highlandsniptechnology.com wrote:
> On Sun, 16 May 2021 22:28:29 +0200, Gerhard Hoffmann <dk4xp@arcor.de>
> wrote:
>
>> Am 16.05.21 um 20:22 schrieb jlarkin@highlandsniptechnology.com:
>>> On Sun, 16 May 2021 20:04:58 +0200, Gerhard Hoffmann <dk4xp@arcor.de>
>>
>>>
>>> It's astounding that you can get a 10 gbps SFP module for $20 from
>>> Amazon. I'd quote about $2000.
>>>
>>> We use them for testing fast o/e converters.
>>
>>
>> SFP is much easier than XFP. Esp. XFP @ 1550 nm.
>> SFP could use VCXLs. That means 5 mA against 50 mA
>> modulation current for DFB, and simple resistors
>> vs. real bias tees in XFP.
>>
>> They could not grow the vertical cavities for the VCXLs
>> for large wavelengths. Maybe it is better now.
>>
>> The AC coupling is a great help, otherwise the
>> modulator would have to provide the DC bias, too,
>> with much larger & slower transistors, and constantly
>> changing the operating point for power and extinction ratio.
>>
>> Infineon sold "us", i.e. Infineon Fiber Optics to Finisar,
>> I spent some weeks with 3 other colleges in San Jose
>> for tech transfer. Me, as a free lancer, not even a real
>> IFX member. :-)
>>
>> At that time, the real cost was about your $2K for XFP. We
>> dreamt of $300 or so. But if you have a _pilot_ production line
>> with 300 people...
>>
>> Gerhard
>>
>
> Even longwave SFPs are cheap now. You can also get WDM ones that
> transmit in both directions over a single fiber.
>
> A genuine Cisco 850nm/10g lists for something like $1200 now.
>
>
> Here's my layout so far.
>
> https://www.dropbox.com/s/2pmf16ng6wpzgo1/Z498A-7.jpg?raw=1
>
> I bailed on the shorted transmission lines to make the short pulse.
> They would have been 16 ohms, and were a nightmare to lay out.
You could make 3x shorted 50 Ohm lines and connect them in parallel. You
might be able to put them on different layers to save area.
Reply by Bill Sloman●May 17, 20212021-05-17
On Monday, May 17, 2021 at 3:35:07 AM UTC+10, jla...@highlandsniptechnology.com wrote:
> I want to make a narrow pulse. This will drive an e/o modulator to be
> an optical "fiducial" marker in a laser system.
>
> I can use a Leo Bodnar style laser driver chip to make a fast edge,
> but I don't know if the chip will propagate a 100 ps wide pulse.
>
> If it won't, maybe I can feed it a step and clip the output with a
> shorted transmission line. Making a 16.66 ohm line is a bit clumsy on
> a PC board... it's pretty fat.
Three 50R transmission lines in parallel would give you 16.67R. Burying a strip-line would gets you a lower impedance line for the same trace width.
The transmission line impedance formulae you find in digital logic application notes tend not be all that useful at 16.6R impedance levels.
I've got a copy of this book
https://www.amazon.com/High-Frequency-Circuit-Design-Measurements-Yip/dp/0412341603
which gives more formulae to cope with a wider range of transmission line impedances.
--
Bill Sloman, Sydney
Reply by ●May 16, 20212021-05-16
On Sun, 16 May 2021 22:28:29 +0200, Gerhard Hoffmann <dk4xp@arcor.de>
wrote:
>Am 16.05.21 um 20:22 schrieb jlarkin@highlandsniptechnology.com:
>> On Sun, 16 May 2021 20:04:58 +0200, Gerhard Hoffmann <dk4xp@arcor.de>
>
>>
>> It's astounding that you can get a 10 gbps SFP module for $20 from
>> Amazon. I'd quote about $2000.
>>
>> We use them for testing fast o/e converters.
>
>
>SFP is much easier than XFP. Esp. XFP @ 1550 nm.
>SFP could use VCXLs. That means 5 mA against 50 mA
>modulation current for DFB, and simple resistors
>vs. real bias tees in XFP.
>
>They could not grow the vertical cavities for the VCXLs
>for large wavelengths. Maybe it is better now.
>
>The AC coupling is a great help, otherwise the
>modulator would have to provide the DC bias, too,
>with much larger & slower transistors, and constantly
>changing the operating point for power and extinction ratio.
>
>Infineon sold "us", i.e. Infineon Fiber Optics to Finisar,
>I spent some weeks with 3 other colleges in San Jose
>for tech transfer. Me, as a free lancer, not even a real
>IFX member. :-)
>
>At that time, the real cost was about your $2K for XFP. We
>dreamt of $300 or so. But if you have a _pilot_ production line
>with 300 people...
>
>Gerhard
>
Even longwave SFPs are cheap now. You can also get WDM ones that
transmit in both directions over a single fiber.
A genuine Cisco 850nm/10g lists for something like $1200 now.
Here's my layout so far.
https://www.dropbox.com/s/2pmf16ng6wpzgo1/Z498A-7.jpg?raw=1
I bailed on the shorted transmission lines to make the short pulse.
They would have been 16 ohms, and were a nightmare to lay out. I hope
the laser driver chip can input and output a 100 ps pulse.
Worst-case, I can dremel a shorted line on a thin substrate and solder
it, sticking up, across the SMA connectors.
--
John Larkin Highland Technology, Inc
The best designs are necessarily accidental.
Reply by Gerhard Hoffmann●May 16, 20212021-05-16
Am 16.05.21 um 20:22 schrieb jlarkin@highlandsniptechnology.com:
> On Sun, 16 May 2021 20:04:58 +0200, Gerhard Hoffmann <dk4xp@arcor.de>
>
> It's astounding that you can get a 10 gbps SFP module for $20 from
> Amazon. I'd quote about $2000.
>
> We use them for testing fast o/e converters.
SFP is much easier than XFP. Esp. XFP @ 1550 nm.
SFP could use VCXLs. That means 5 mA against 50 mA
modulation current for DFB, and simple resistors
vs. real bias tees in XFP.
They could not grow the vertical cavities for the VCXLs
for large wavelengths. Maybe it is better now.
The AC coupling is a great help, otherwise the
modulator would have to provide the DC bias, too,
with much larger & slower transistors, and constantly
changing the operating point for power and extinction ratio.
Infineon sold "us", i.e. Infineon Fiber Optics to Finisar,
I spent some weeks with 3 other colleges in San Jose
for tech transfer. Me, as a free lancer, not even a real
IFX member. :-)
At that time, the real cost was about your $2K for XFP. We
dreamt of $300 or so. But if you have a _pilot_ production line
with 300 people...
Gerhard
Reply by ●May 16, 20212021-05-16
On Sun, 16 May 2021 20:04:58 +0200, Gerhard Hoffmann <dk4xp@arcor.de>
wrote:
>Am 16.05.21 um 19:34 schrieb jlarkin@highlandsniptechnology.com:
>>
>> I want to make a narrow pulse. This will drive an e/o modulator to be
>> an optical "fiducial" marker in a laser system.
>>
>> I can use a Leo Bodnar style laser driver chip to make a fast edge,
>> but I don't know if the chip will propagate a 100 ps wide pulse.
>
>I see no problem with a 10 GBit/sec laser driver. A bit cell is 100 ps,
>and they must switch on/off in this time and present a nice open eye
>to the laser. And the laser will produce overshoot itself when it was
>quite off before. IIRC, the extinction ratio was expected to be > 10 dB.
I'm thinking that the laser driver will propagate a 100 ps wide pulse,
and I can make that. I might lay out a test board with the tx lines
such that I can Dremel them away as needed.
The annoyance is that everybody who makes fast stuff seems to assume
telecom applications, ac coupled balanced fast data. So you have to
experiment to see how/if they work dc coupled for pulses.
>
>> If it won't, maybe I can feed it a step and clip the output with a
>> shorted transmission line. Making a 16.66 ohm line is a bit clumsy on
>> a PC board... it's pretty fat.
>
>Hey, wasn't it you who complained about the thin FR4 on the
>Chinese 4 layer boards?
Yes, that was unexpected. Bad for fast stuff. Every pad become a
capacitor.
>
>
>BTW. we used FR4 for our 10 GB/s XFP transceivers, and a Kapton tape
>to connect the TO-52 TOSA. That tape was a modeling nightmare in HFSS.
>
>OK, with a e/o modulator it will be simpler than feeding the laser directly.
The e/o also does shaping of a light pulse that's furnished. The
fiducial is an occasional calibration thing that we will mix in.
If all we wanted was a fast light spike, we might use a gain-switched
laser.
It's astounding that you can get a 10 gbps SFP module for $20 from
Amazon. I'd quote about $2000.
We use them for testing fast o/e converters.
>
>:-) Gerhard
--
John Larkin Highland Technology, Inc
The best designs are necessarily accidental.
Reply by Gerhard Hoffmann●May 16, 20212021-05-16
Am 16.05.21 um 19:34 schrieb jlarkin@highlandsniptechnology.com:
>
> I want to make a narrow pulse. This will drive an e/o modulator to be
> an optical "fiducial" marker in a laser system.
>
> I can use a Leo Bodnar style laser driver chip to make a fast edge,
> but I don't know if the chip will propagate a 100 ps wide pulse.
I see no problem with a 10 GBit/sec laser driver. A bit cell is 100 ps,
and they must switch on/off in this time and present a nice open eye
to the laser. And the laser will produce overshoot itself when it was
quite off before. IIRC, the extinction ratio was expected to be > 10 dB.
> If it won't, maybe I can feed it a step and clip the output with a
> shorted transmission line. Making a 16.66 ohm line is a bit clumsy on
> a PC board... it's pretty fat.
Hey, wasn't it you who complained about the thin FR4 on the
Chinese 4 layer boards?
BTW. we used FR4 for our 10 GB/s XFP transceivers, and a Kapton tape
to connect the TO-52 TOSA. That tape was a modeling nightmare in HFSS.
OK, with a e/o modulator it will be simpler than feeding the laser directly.
<
https://www.flickr.com/photos/137684711@N07/42256986280/in/album-72157662535945536/lightbox/
>
:-) Gerhard
Reply by ●May 16, 20212021-05-16
I want to make a narrow pulse. This will drive an e/o modulator to be
an optical "fiducial" marker in a laser system.
I can use a Leo Bodnar style laser driver chip to make a fast edge,
but I don't know if the chip will propagate a 100 ps wide pulse.
If it won't, maybe I can feed it a step and clip the output with a
shorted transmission line. Making a 16.66 ohm line is a bit clumsy on
a PC board... it's pretty fat.
Version 4
SHEET 1 880 680
WIRE 96 128 -16 128
WIRE 320 128 176 128
WIRE 448 128 320 128
WIRE -16 176 -16 128
WIRE 288 176 256 176
WIRE 448 176 448 128
WIRE 288 208 288 176
WIRE 320 208 320 128
WIRE -16 288 -16 256
WIRE 256 320 256 176
WIRE 288 320 288 304
WIRE 288 320 256 320
WIRE 320 320 320 304
WIRE 320 320 288 320
WIRE 448 320 448 256
WIRE 256 352 256 320
FLAG 256 352 0
FLAG -16 288 0
FLAG 448 320 0
SYMBOL res 192 112 R90
WINDOW 0 70 59 VBottom 2
WINDOW 3 75 60 VTop 2
SYMATTR InstName Rd
SYMATTR Value 25
SYMBOL voltage -16 160 R0
WINDOW 0 -65 130 Left 2
WINDOW 3 -95 171 Left 2
WINDOW 123 0 0 Left 0
WINDOW 39 0 0 Left 0
SYMATTR InstName Vd
SYMATTR Value PULSE(0 1 100p 25p 25p 1)
SYMBOL tline 304 256 R90
WINDOW 0 -39 90 VTop 2
WINDOW 3 24 155 VBottom 2
SYMATTR InstName T1
SYMATTR Value Td=50p Z0=16.7
SYMBOL res 432 160 R0
WINDOW 0 -56 51 Left 2
WINDOW 3 -54 80 Left 2
SYMATTR InstName R2
SYMATTR Value 50
TEXT -224 272 Left 2 !.tran 1n
TEXT -264 144 Left 2 ;Shorted Txline
TEXT -288 176 Left 2 ;Fiducial Generator
TEXT -272 224 Left 2 ;JL May 16 2021
--
John Larkin Highland Technology, Inc
The best designs are necessarily accidental.