Looking for Cheap, Ultra Fast LEDs (2-3ns)

[supperman]

so I think I'm happy with LEDs giving me around 400 Mbit... I think it should be possible..

I think you could do a 4Mb/s channel for $0.50, but I think there's no chance of 400Mb/s.

Not sure where you are coming from. I'm running at 3ns switching times now.. and am looking for help finding other LEDs to test. I am of course open to other circuits as well.. but most of what I see online are inferior at this point.

[supperman]

Did you see the speedup circuit in the paper? That should fit in your budget and might eek out some extra performance from indicator LEDs if necessary.

Yes, I have seen this and decided to go a different way. I find that it is limited in how I can tweak it.. and it does not run efficiently for continuous data. (As I think they even point out in their paper). I don't use capacitors or inductors.. I use separate circuits  to drive pos and neg spikes and have complete control over duration and current on them.

when you look at LED data sheets you can quickly see that continuous drive current may be 70ma.. but they can handle spikes up to an AMP. The above circuit can't deliver that.

[supperman]

This sounds a lot like an x-y problem.

Incidentally, getting lots (>>gigabit) data from A-to-B for cheap is the field I'm working in, so perhaps I could help with some outside-of-the-box ideas if you can give more information/limits you are facing. What volume are you looking at?

Depending on what output type you have, you might be able to do something with larger symbol size? instead of OOK, consider 4 or even 8 ASK to lower symbolrate?

I'm not sure I know what you are talking about? Encoding data for error correction? I'm not there yet.. just doing single channel bit-rate speed tests. I'll take the problems  as they come.. but distances are short and I'll layout all paths with matched lengths so I'm hoping that data will be reasonably in sync and that I don't need to worry. It is video so 100% lossless is also not required.. and I'm hoping to sit on top of some protocol used by the FPGA to serialize and combine..

[skylar]

I'm using it to send data across about a 1 cm air gap.. and I use around 10+ of them to get the data I need. Not easy.. since we are talking about serious data over very cheap components.

I always have to remind myself that light travels about one foot at 1ns.. to get some perspective..

Also - running this with an FPGA which is limited to about 600 Mbit per pin so taking it down to say a single laser would add all sorts of cost to serialize multiple channels.. etc.. so I think I'm happy with LEDs giving me around 400 Mbit... I think it should be possible..

Is the 1cm air gap critical or just part of this particular LED implementation?

A single SERDES pair and an AC coupled differential pair is about the same cost as 10x of your LED circuits and is good for our isolation needs and certainly meets your speed requirements. For better isolation, the guts of a GigE SFP provides the optical link in our stuff (get a FC 4Gb since you need the speed, they even go up to >16Gb these days). Cost wise you could probably buy one and split the Tx and Rx side onto your own boards with only a little tweaking. Overall it does probably double your parts cost, but its proven tech :/

[supperman]

I'm using it to send data across about a 1 cm air gap.. and I use around 10+ of them to get the data I need. Not easy.. since we are talking about serious data over very cheap components.

I always have to remind myself that light travels about one foot at 1ns.. to get some perspective..

Also - running this with an FPGA which is limited to about 600 Mbit per pin so taking it down to say a single laser would add all sorts of cost to serialize multiple channels.. etc.. so I think I'm happy with LEDs giving me around 400 Mbit... I think it should be possible..

Is the 1cm air gap critical or just part of this particular LED implementation?

A single SERDES pair and an AC coupled differential pair is about the same cost as 10x of your LED circuits and is good for our isolation needs and certainly meets your speed requirements. For better isolation, the guts of a GigE SFP provides the optical link in our stuff (get a FC 4Gb since you need the speed, they even go up to >16Gb these days). Cost wise you could probably buy one and split the Tx and Rx side onto your own boards with only a little tweaking. Overall it does probably double your parts cost, but its proven tech :/

Yes, there is unfortunately a physical requirement - air gap AND about 10+ units... so I have to work with that.
Also.. if I could use a single link.. now we are talking about an FPGA with 2x 5 gbit ports - or a dedicated chip to do the same.. and that certainly blows the budget, right? I have not gotten FPGA quotes with high speed ports.. but adding them seems to triple the price at least at retail.

[skylar]

Yes, there is unfortunately a physical requirement - air gap AND about 10+ units... so I have to work with that.
Also.. if I could use a single link.. now we are talking about an FPGA with 2x 5 gbit ports - or a dedicated chip to do the same.. and that certainly blows the budget, right? I have not gotten FPGA quotes with high speed ports.. but adding them seems to triple the price at least at retail.

I was talking about a dedicated chip, the SERDES (serial/de-serializer) I find in your ~10bit 4Gb use case are about $6/EA and let you use the same 10x 400Mb ports on your FPGA. Since you mentioned its a video application, there's even dedicated 16bit RGB video SERDES like the Renesas ISL76322ARZ that popped up in my search, not sure what you are doing with your FPGA but that's an interesting option to just shuttle video data around that may not need an FPGA?

[dmendesf]

He was talking about increasing the data rate without increasing the simbol rate by coding more than one bit per simbol. Suppose instead of on / off you had 4 different brighteness levels... Then you could double the data rate without needing a faster diode. There are other coding schemes to achieve this.

[supperman]

He was talking about increasing the data rate without increasing the simbol rate by coding more than one bit per simbol. Suppose instead of on / off you had 4 different brighteness levels... Then you could double the data rate without needing a faster diode. There are other coding schemes to achieve this.

Ah.. Understood! Not possible in this design since the brightness will vary based on external forces (i.e. LED and sensor will not always be aligned perfectly) I think I can safely clip on "on / off" but can get no more detail out of it.

[supperman]

Yes, there is unfortunately a physical requirement - air gap AND about 10+ units... so I have to work with that.
Also.. if I could use a single link.. now we are talking about an FPGA with 2x 5 gbit ports - or a dedicated chip to do the same.. and that certainly blows the budget, right? I have not gotten FPGA quotes with high speed ports.. but adding them seems to triple the price at least at retail.

I was talking about a dedicated chip, the SERDES (serial/de-serializer) I find in your ~10bit 4Gb use case are about $6/EA and let you use the same 10x 400Mb ports on your FPGA. Since you mentioned its a video application, there's even dedicated 16bit RGB video SERDES like the Renesas ISL76322ARZ that popped up in my search, not sure what you are doing with your FPGA but that's an interesting option to just shuttle video data around that may not need an FPGA?

Sorry - in my mind I thought you meant optical transceiver..  which of course I would still need.. at a high speed rate. I don't know of anything cheap here.. $10 for the transmitter, $10 for the receiver? Another $6 for 2nd SERDES.. optics? stuff adds up very quick. And debugging 4gbit signals.. am limited to 1ghz scope.

I'm not saying no. perhaps you have links to come components here that simplify things? I just don't know of a transmitter or receiver that directly takes/sends a differential signal at that speed that does not cost an arm and a leg... and if I'm building one I think I may be in a world of hurt at that data rate..

This is how I always come back to the LED design

(and for the people paying attention.. yes, potentially there is a way to use a single transmit/receive rather than 10+. It is either or)

[StillTrying]

so I think I'm happy with LEDs giving me around 400 Mbit... I think it should be possible..

I think you could do a 4Mb/s channel for $0.50, but I think there's no chance of 400Mb/s.

Not sure where you are coming from. I'm running at 3ns switching times now..

If you're putting a digital 100Mb/s in one end and recovering a 100Mb/s digital output from the other I've no idea how you're doing that with a few $0.10 components!

and am looking for help finding other LEDs to test.

I think if any one here knows much about switching normal LEDs and photodiodes at these speeds they're keeping it to themselves.

[TheUnnamedNewbie]

He was talking about increasing the data rate without increasing the simbol rate by coding more than one bit per simbol. Suppose instead of on / off you had 4 different brighteness levels... Then you could double the data rate without needing a faster diode. There are other coding schemes to achieve this.

Ah.. Understood! Not possible in this design since the brightness will vary based on external forces (i.e. LED and sensor will not always be aligned perfectly) I think I can safely clip on "on / off" but can get no more detail out of it.

Usually you would solve this with a low-pass filter. Using encoding like 8b/10b or 64b/66b, you can actually remove DC from any datastream (and thus you have a signal that has no spectral content below a certain frequency). You then block that out from your signal with a low-pass filter (usually in the form of a DC block). Since environmental changes will be low frequency (even if your device was spinning at 1000 RPM, it would be a low-frequency change with respect to your >100 Mbit/s signal) it would be stopped by this DC block.

[supperman]

Usually you would solve this with a low-pass filter. Using encoding like 8b/10b or 64b/66b, you can actually remove DC from any datastream (and thus you have a signal that has no spectral content below a certain frequency). You then block that out from your signal with a low-pass filter (usually in the form of a DC block). Since environmental changes will be low frequency (even if your device was spinning at 1000 RPM, it would be a low-frequency change with respect to your >100 Mbit/s signal) it would be stopped by this DC block.

Ok.. this does sound interesting! Of course these transitions are low frequency.. and I did not consider them as such. You got me thinking in a new way.. so thank you. Let me dig a little deeper and perhaps get back to you with questions. Any links would be helpful if you have them.

I'm still rather new to this forum.. but I have to say.. it has been a positive experience.

[skylar]

Sorry - in my mind I thought you meant optical transceiver..  which of course I would still need.. at a high speed rate. I don't know of anything cheap here.. $10 for the transmitter, $10 for the receiver? Another $6 for 2nd SERDES.. optics? stuff adds up very quick. And debugging 4gbit signals.. am limited to 1ghz scope.

I'm not saying no. perhaps you have links to come components here that simplify things? I just don't know of a transmitter or receiver that directly takes/sends a differential signal at that speed that does not cost an arm and a leg... and if I'm building one I think I may be in a world of hurt at that data rate..

As you say, you would need an (or 2x) optical transceiver for your air gap. I luckily have gotten these as scrap or cheap used components as the SANs out there go to the higher data rates and the 4/8Gb transceivers are no longer as dear.

[supperman]

Usually you would solve this with a low-pass filter. Using encoding like 8b/10b or 64b/66b, you can actually remove DC from any datastream (and thus you have a signal that has no spectral content below a certain frequency). You then block that out from your signal with a low-pass filter (usually in the form of a DC block). Since environmental changes will be low frequency (even if your device was spinning at 1000 RPM, it would be a low-frequency change with respect to your >100 Mbit/s signal) it would be stopped by this DC block.

I guess the first question is how would you modulate the signal.. with AtoD and DtoA converters? I can see avoiding DtoA via some smart ways of powering the LED.. but on the receiver side it seems unavoidable? Sounds expensive at that data rate... or is there some smart way of using cheap compareitors ?

[Marco]

Fast ADCs are ridiculously overpriced ... maybe PPM/PWM can get you a little more out of a single channel.

[technix]

What for though? Is it ultra-fast display? Fiber optics?

If it is the former, I have nothing. If it is the latter, maybe you can use a ready-made module (TOSLINK transmitter for slower ones, SFP+ module for something that is real fast) and just focus on the digital logic part. If your source chip has serdes, it is likely you can get SFP+ would work with that.

[supperman]

Fast ADCs are ridiculously overpriced ... maybe PPM/PWM can get you a little more out of a single channel.

Agree.. dug a little deeper.. like $3.5 per channel at 8 bit and 100msample. One could probably get 2-3 bits per sample of data sampling both amplitude and frequency shifts (amplitude at the crossing point) on a carrier wave.... so 400mbit+ but it really busts the budget.

[supperman]

What for though? Is it ultra-fast display? Fiber optics?

If it is the former, I have nothing. If it is the latter, maybe you can use a ready-made module (TOSLINK transmitter for slower ones, SFP+ module for something that is real fast) and just focus on the digital logic part. If your source chip has serdes, it is likely you can get SFP+ would work with that.

This is for video transmission in a very specialized "air gaped" setup. I would just call it video.. not super fast or anything.. more than 1080p, less than 4k.. uncompressed.

[dmendesf]

If you want to quantize in 4 steps (to double your data rate) you could try using the embedded LVDS receivers in the receiving FPGA as comparators. Crude but cheap. Of course this won't scale much more than that.

[Marco]

I think if any one here knows much about switching normal LEDs and photodiodes at these speeds they're keeping it to themselves.

The standard fast'ish cheap photodiode is SFH2701 ... the Chinese must have low cost faster photodiodes though, they sell Fiber->CATV converters for a couple bucks.

PS. actually they seem to be cheap ROSAs, not loose photodiodes. Less useful.

[supperman]

I think if any one here knows much about switching normal LEDs and photodiodes at these speeds they're keeping it to themselves.

The standard fast'ish cheap photodiode is SFH2701 ... the Chinese must have low cost faster photodiodes though, they sell Fiber->CATV converters for a couple bucks.

PS. actually they seem to be cheap ROSAs, not loose photodiodes. Less useful.

Yes, that is the one I'm using.  2ns rise and fall for $0.22 at DigiKey in volume..

[technix]

What for though? Is it ultra-fast display? Fiber optics?

If it is the former, I have nothing. If it is the latter, maybe you can use a ready-made module (TOSLINK transmitter for slower ones, SFP+ module for something that is real fast) and just focus on the digital logic part. If your source chip has serdes, it is likely you can get SFP+ would work with that.

This is for video transmission in a very specialized "air gaped" setup. I would just call it video.. not super fast or anything.. more than 1080p, less than 4k.. uncompressed.

If it is video, maybe HDBaseT over 10Gbps fiber?

[StillTrying]

Has the OP managed to make a $0.50 400Mb/s opto isolator yet. 

[supperman]

Has the OP managed to make a $0.50 400Mb/s opto isolator yet. 

I'm around 200mbit and around $1 for sender and receiver components. Still some stability issues. I'm optimistic to raise this speed up into the target range. Mostly I'm dealing with radiation.. so it is as much about PCB design as anything else. This of course does not count the two FPGAs on either end that have some useful i/o capabilities.

Again, I'm not here to debate if what I'm already doing can be done or not. I was asking about other LED components that may work better than what I got.. and it has lead to some interesting conversations.

[StillTrying]

Mostly I'm dealing with radiation.. so it is as much about PCB design as anything else.

I hope that wasn't a surprise!

it has lead to some interesting conversations.

We don't need to know why/what the 1cm air gap is for, and you've given reasons why some solutions aren't useful, but without knowing what/if anything is useful I think some of us will be finding the conversations just frustrating.