Floating probe! For $2.50

[StillTrying]

Well I'm glad the anti-phase optos seem to work in keeping the linearity without any feedback, because AFAIK that was untested in practice.

But, (there's always a but ) have you seen my photo diode amp step response at the same 500ns/Div speed. LEDphotoTRopamp.gif

"the more BW the better!"

Yep, it's quite possible to get it much higher than that, I'm setting the target at 1.5M+.

I simulated the 5 transistor long tailed pair arrangement in the end. The problem is the high 2k5 impedance at the bases, it's much too high, together with the PD capacitance it limits the BW at that point to ~600kHz.
Everybody will know more about that 5 TR arrangement than me, so is it possible to get the base impedance down to a few 100 R. After about an hour of trying I gave up and drew this 1 transistor instead.

I've got the impedance at the base ~160R, and the BW down to 3MHz, because I think that might be about the best the opto can do.
The input current variation is 7mA, -3.5mA to +3.5mA, and the collector OUTput 138mV pk-pk, I've doubled the frequency of my test waves to 200kHz, and they still look good!

A problem with that simple 1 transistor is that the output ~2V DC level is not very well defined and varies a bit with temp.
I'll add the LT.asc and hopefully someone will suggest simple improvements.

[David Hess]

The 4N35 and similar transistor output optocouplers which provide access to the base connection are a lot faster if the base connection is used.

I suspect for a linear application it might be best to treat the output transistor as a photodiode driving a transimpedance amplifier.

[BrianHG]

Well I'm glad the anti-phase optos seem to work in keeping the linearity without any feedback, because AFAIK that was untested in practice.

But, (there's always a but ) have you seen my photo diode amp step response at the same 500ns/Div speed. LEDphotoTRopamp.gif

"the more BW the better!"

Yep, it's quite possible to get it much higher than that, I'm setting the target at 1.5M+.

I simulated the 5 transistor long tailed pair arrangement in the end. The problem is the high 2k5 impedance at the bases, it's much too high, together with the PD capacitance it limits the BW at that point to ~600kHz.
Everybody will know more about that 5 TR arrangement than me, so is it possible to get the base impedance down to a few 100 R. After about an hour of trying I gave up and drew this 1 transistor instead.

I've got the impedance at the base ~160R, and the BW down to 3MHz, because I think that might be about the best the opto can do.
The input current variation is 7mA, -3.5mA to +3.5mA, and the collector OUTput 138mV pk-pk, I've doubled the frequency of my test waves to 200kHz, and they still look good!

A problem with that simple 1 transistor is that the output ~2V DC level is not very well defined and varies a bit with temp.
I'll add the LT.asc and hopefully someone will suggest simple improvements.

Place compensation caps across R1 and R2, maybe around 100pf.  You will improve bandwidth further.
Change that BC546 to a MPSH10 or MMBTH10LT1.
Also, Toshiba's optocoupler TLP2719 is around 10x faster than the 4N25.

[JS]

  Well, I got -3dB at 550kHz but with some ringing.. +0.8dB at 440kHz. I can observe the AC component on the ringing in the virtual earth node, so I think here is the NE5532 not being able to keep up, I could try an NE5534 which is a bit faster but I'm not sure if it's going to be stable in this config without external compensation.




  Compensating with 200pF still doesn't solve the ringing, 270pF completly eliminates it but with a BW limit of 280kHz. Using a 220pF seems to be the sweet spot with very little ringing and the -3dB point is at 350kHz. Here is the response with the 220pF cap with some series resistance, -3dB point a bit over 350kHz.


JS

[BrianHG]

2x J113, or J112 jfet in a push pull source follower/constant current configuration replacing the TLO81 opamp and a bias adjustment on R1 and R2's values could improve bandwidth and get rid of some ring.

[David Hess]

The transistors acting as photodiodes add shunt capacitance at the inverting input of the NE5532 operational amplifier destabilizing it.  A capacitor across the feedback resistor cancels the shunt capacitance but also lowers bandwidth.

There are application notes about how to improve performance of that circuit configuration.  Search for photodiode transimpedance amplifiers.

[StillTrying]

-3dB point a bit over 350kHz.

With NE5532s in simulation that's the about the best I could get, with NE5532 and opto I think that might be the limit.

I've speed tested a cny1711 opto, just it's base-collector junction. It's over 10 times slower than a separate LED and photo diode, even a white LED + PD is 10 times faster than the opto.

[JS]

Thanks for the ideas, I'll try them out when I find some time on my hands...

-3dB point a bit over 350kHz.

With NE5532s in simulation that's the about the best I could get, with NE5532 and opto I think that might be the limit.

I've speed tested a cny1711 opto, just it's base-collector junction. It's over 10 times slower than a separate LED and photo diode, even a white LED + PD is 10 times faster than the opto.

Are you suggesting discrete LED and photodiode are faster than the collector base junction as photodiode?

JS

[BrianHG]

Thanks for the ideas, I'll try them out when I find some time on my hands...

-3dB point a bit over 350kHz.

With NE5532s in simulation that's the about the best I could get, with NE5532 and opto I think that might be the limit.

I've speed tested a cny1711 opto, just it's base-collector junction. It's over 10 times slower than a separate LED and photo diode, even a white LED + PD is 10 times faster than the opto.

Are you suggesting discrete LED and photodiode are faster than the collector base junction as photodiode?

JS

Yes, take a look at the spec on this photodiode: https://www.digikey.ca/product-detail/en/finisar-corporation/P850-2124-001/P850-2124-001-ND/4416177

It has a rise and fall time of 30 PICOSECONDS, however, it is pricey at 23$.  This makes an optocoupler look like watching paint dry by comparison.

Ok, for a reasonable priced photodiode: https://www.digikey.com/product-detail/en/osram-opto-semiconductors-inc/SFH-2701/475-2967-1-ND/2794398

At 87cents, and a 2ns rise and fall time, it makes the optocoupler look like a snail speed device.  And there are 50k units in stock.

I'm sorry to say, but, 2 blue leds, or 2 high speed 820nm IR leds, with 2 of these 87cent photodiodes in a push pull drive can reach beyond 200mhz bandwidth with a proper discrete pre-amp on the IR led, and similar for the photodiode.

However, you will need to mount them in a small black straw/heat shrink tube.  Though, with a little space, a narrow beam blue LED, you can now have 10s of KV isolation.

You might make it to 500mhz bandwidth with a laserdiode...

[StillTrying]

Are you suggesting discrete LED and photodiode are faster than the collector base junction as photodiode?

Well that's what I've found in one opto test in practice. In simulation with the 4N25 I can get passed the 1 - 2us speed limit by using just the base-collector photo diode part and get the analogue BW up to 10M. But in practice with a cny1711 the 1-2us speed limit seems to be on the b-c photo diode itself, even 8V reverse bias made no noticeable difference to the rise time.
I had to make the opto's pulses 5us and 10us wide to match the shape of the 0.5us and 1us pulses from a 5mm super bright white LED and SFH213 PD - so opto's PD was ~10X slower!

It mostly matches what you've found, a simulated BW of 2M becomes closer to 200K in practice, although I still think the NE5532 is a limit.

A super bright white ~1MHz looks the easiest ATM, it might have 250ns rise and fall times but it reaches the flat top of the 1us pulse before the red, and is 10-100X brighter. A super bright green, red or cool white might be faster, but I don't have any loose ones to test - yet.

I'll probably test the linearity of the super bright white around 3mA - 30mA soon. I'm expecting the open loop linearity to be very good, and the BW ~1MHz.

The blue trace is the 1us through a super bright white.

[BrianHG]

A super bright white ~1MHz looks the easiest ATM, it might have 250ns rise and fall times but it reaches the flat top of the 1us pulse before the red, and is 10-100X brighter. A super bright green, red or cool white might be faster, but I don't have any loose ones to test - yet.

Super bright white is a blue led with yellow phosphor to make it look white.  The super bright blue should be the fastest as it has no phosphor and it's a fundamental color generated by the die directly.  A true super bright 405nm violet led or UV ones should be fastest.

[StillTrying]

"I'll probably test the linearity of the super bright white around 3mA - 30mA soon. I'm expecting the open loop linearity to be very good, and the BW ~1MHz"

At 8.5mA +/-5mA the super bright white LEDs light output seems very linear.

Max. input(and output) is 200mVpp, with the ca3140 10%-90% rise/fall times ~250ns, -1dB at 280kHz so -3dB is going to be <1MHz.

Green trace is the light output.

[BrianHG]

"I'll probably test the linearity of the super bright white around 3mA - 30mA soon. I'm expecting the open loop linearity to be very good, and the BW ~1MHz"

At 8.5mA +/-5mA the super bright white LEDs light output seems very linear.

Max. input(and output) is 200mVpp, with the ca3140 10%-90% rise/fall times ~250ns, -1dB at 280kHz so -3dB is going to be <1MHz.

Green trace is the light output.

What does the throughput of a triangle wave look like?

[StillTrying]

"What does the throughput of a triangle wave look like?"

I don't know, my simple AF SG doesn't do triangular, but as there's already a CMOS 555 on the other end of the breadboard, I might produce and capture some rough triangular before swapping the SB white. I'd expect the linearity to be OK, but the different rise and fall times of the SB white to become visible around 10kHz.

The first start of clipping on the bottom of sines was the 3140 struggling to pull down against the 1k2 pull-up (which was there help it!), rather than the LED current going <3mA. By simply removing the 1k2 and letting the 3140 supply all the LED current the dynamic range before clipping increased ~240mV to ~350mVpp, and the speed, with a slight increase of the 8.5mA the SB white BW must now be close to 1MHz, probably about the best a white LED is going to get at these low LED currents.

With a -ve supply availiable, an NE5532's higher output current and BW might be quite good there, the voltage gain needed is very low because the voltage across the LED only varies ~100mV + the 20R.

SuperBright Green, Orange and Blue's have arrived, it might be a day or 2 before I test them.

[BrianHG]

According to these guys, bandwidth tests done, blue really the best at 800mb, then green, then red:
http://www.spie.org/newsroom/3941-optical-wireless-network-built-on-white-light-leds-reaches-800mb/s
The problem with red is that there are so many different formulations, from 620nm through 670nm, some slow, some fast.  You can at least count on super bright blue always being the same formulation.

[IDEngineer]

According to these guys, bandwidth tests done, blue really the best at 800mb, then green, then red

Hmm, I'm missing something in that article. They say "The maximum data rates at which the BER did not exceed a 2·10−3 forward error correction threshold were ~ 294 (red), ~ 223 (green), and ~ 286 (blue) Mb/s", meaning that red was fastest and green was slowest. Granted these were three dies on an RGB LED, and they were using a modulation scheme to push the bit rate past the raw LED bandwidth, but presuming the same tests were run on all three dies it appears red was the fastest. I honestly *expected* blue to be the fastest... what am I missing?

[BrianHG]

According to these guys, bandwidth tests done, blue really the best at 800mb, then green, then red

Hmm, I'm missing something in that article. They say "The maximum data rates at which the BER did not exceed a 2·10−3 forward error correction threshold were ~ 294 (red), ~ 223 (green), and ~ 286 (blue) Mb/s", meaning that red was fastest and green was slowest. Granted these were three dies on an RGB LED, and they were using a modulation scheme to push the bit rate past the raw LED bandwidth, but presuming the same tests were run on all three dies it appears red was the fastest. I honestly *expected* blue to be the fastest... what am I missing?

That's because they are trying to communicate with overall white light so your office LED lighting is the network and people wont be working in an all blue lit environment.  Look at the first table with the axis " Demonstrated rate [Mbps] ".  You see on that table, the red is the slowest while the blue is 800Mbps all on it's own.  Get rid of all other light colors in you office and illuminate it pure blue, and you will get 800Mbps all in one led, one wavelength of light.

If you want your employees the luxury of white lighting, mixing 3 colors to always achieve a single brightness and color of white, WHILE TRANSMITTING DATA, then the figures they quote for their trick to keep one brightness of white with one color temperature are the maximum speeds they can achieve.

Pure white LEDs are slower due to them being blue LEDs with a yellow phosphor which has an inherent afterglow to it slowing down the maximum Mbps.

[StillTrying]

I joined a string of LEDs and with 5mA through the string, measured the uAs produced by a SFH213 in contact with the LED, they're the highest blue and green uAs I've seen.
uA
235 SB Blue
225 SB Green
140 Warm White
080  Medium White
070  SB Red
050  SB Orange
026  Std Red
025  old IR

With the SB blue or SB green the BW now seems to limited by what the 3140's output/top of LED can do, if I scope the 3140's output, the voltage shape is a close match to the received light shape.

[BrianHG]

I joined a string of LEDs and with 5mA through the string,

What doe this mean?
Are you driving all these LEDs in series?  Parallel?
Different LEDs have different capacitance and current rise and fall times based on it's internal switching time.  Wouldn't testing 1 led at a time yield better results.

[StillTrying]

"What does this mean?
Are you driving all these LEDs in series?  Parallel?"

5mA DC from a PSU through the series string, and a bare SFH213 connected directly to DMM to measure the max. head on uA photo current from each 5mm water clear LED, 3 times around the string and averaged, so not exact uA, (depends on the exact head on alignment), the 5mA might have changed a bit, but the order of sensitivity was the same for each of the 3 times the string was measured.

I'm surprised no one asks what I'm using to receive the light, most of the time I'm using the PD +1 transistor from the light bulb thread, I'd add the imperfect 1 TR receive on to the schem in the previous post #42 when I find it.

[StillTrying]

I tried a 5mm UV LED, it's light output had horrible curves lasting microseconds.

https://www.eevblog.com/forum/chat/20w-halogen-bulb-viewed-by-a-photodiode/?action=dlattach;attach=547853;image