Re: Twin T circuit wanted

How do you do that? Summing on/off light sources? I guess the question is, how small?

John

Low distortion sine wave + DC driving IR LED PD goes to TIA with x64 AC gain and x1 DC gain Auxiliary light source (fibre coupled laser in this case) provides CW light Bridge circuit (like an old HP THD meter) to null out sine wave and improve dynamic range

Set up IR LED to give some decent AC photocurrent like 5 uA, null carefully, move laser in and out while changing PD reverse bias voltage.

Look at residual on digital scope with FFT.

This allows me to see nonlinearities down to about -70 dB. It could go further, but the silly laser mode hops like mad, and it takes forever to integrate it away.

Interim conclusion: InGaAs is a lot less linear than Si. The quantum efficiency improves by something like 5% for bias levels between 0 and 5 V.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal
ElectroOptical Innovations
55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058
hobbs at electrooptical dot net
http://electrooptical.net

So you are sort of summing light sources. I guess.

If you measure the magnitude of the AC component of the PD signal, at various points of added laser light, you get the slope gain at various points, and the distortion of the LED doesn't matter. Is that the idea?

LEDs are pretty limear at higher currents, but I don't know about 70 dB. Actually, it wouldn't be all that hard to find out.

Does that imply amplitude nonlinearity at fixed bias?

What about temperature? Maybe the bias voltage needs a TC correction.

John

Exactly. Just like a two-tone tester, where you sum the tones right at the output to avoid IMD internal to the tester.

Yes. That measures the slope rather than the actual curve, which is what I'd really prefer in this instance.

Yes, and it's worst at zero bias--maybe 2x or so, in the range I care about. Interestingly the AC output signal shows a straight-line growth with bias--effectively the quantum efficiency goes up about 5% from 0 to

5V bias.

I hope not--this gizmo is going to have to work in a fairly harsh environment eventually. Today I'm going to do an eyeball fit to the distortion polynomial, to see if I can predict how much the output gets compressed at higher powers. It's on the order of 2% at 100 uW in a

300-um diameter detector at zero bias, which is really stinky compared with silicon.

And InGaAs is the best of the infrared detector materials. Infrared is just a lot harder than visible, for a whole bunch of reasons.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal
ElectroOptical Innovations
55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058
hobbs at electrooptical dot net
http://electrooptical.net

How about using N led's, minimum two.

Turn them on and off in all possible states, measure PD current with a very good meter, and analyze the results. See how well things sum that ought to sum.

You could also trim the individual LED currents. It would be reasonable to do a series of experiments that would precisely adjust the LED light outputs to be equal, or to be precise 2:1 (optical binary DAC) steps. Then you can play with combinations. LED self-heating would be the biggest error, so something clever would need to be done about that.

That's the sort of thing that would be fun to do as a product, if you thought that anybody would buy it.

Knowing nothing about semiconductors doesn't keep me from speculating. Maybe the high defect density causes recombinations at low drift velocities.

At any rate, I should keep my PD power supplies stiff.

John

The problem is that the photodiode is only 300 um in diameter, so there's a limit to how many LEDs you can crowd in there. I only had an afternoon to design and wire up a tester, so I used what I had lying around. My ABQ customer is a great outfit, but they have zilch prototyping supplies, so I built it at home.

It's so SMT here, I was pathetically grateful to find a Radio Shack a mile away. I can do dead-bug with 0805s, but anything smaller is hard. Next time I'll bring a box of parts as well as a bag of tools. (I brought my own micromanipulators.)

Back in the palmy days, I built a little tester for transistor log conformity that worked like that--it had a 10-turn pot controlling the collector current, with a switching x1-x2 amp. It used an LTC1043 low charge injection analog switch, which stored Vbe at the x1 setting and subtracted it from Vbe at the x2 setting. Looking at how the difference varied with collector current for different devices was very illuminating.

If people knew that it was a problem, they would--but they generally have no idea, and radiometric calibration is hard to do to much better than 1%. It's mind-boggling, the number of people who wire up a photodiode and expect it to be linear to all 98 bits of their shiny new delta-sigma. So it's the sort of tester that (ideally) allows you to ship other magic products, if you're sufficiently clueful. Despite my best efforts to educate folks, a great many just wave a dead chicken over their PD and TIA, and hope that it works. Then they either sweep it under the rug or redefine what they mean by 'working'. :(

Yep. But it's primarily quantum efficiency that changes with bias, not the linearity--anywhere above half a volt or so, bias doesn't make much difference to the linearity vs photocurrent, at least not down in the sub-100uA range where I'm working. The output vs bias curves are mostly parallel straight lines, with some droop at zero bias and higher photocurrent.

It's apparently important to illuminate the photodiode from the P-side.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal
ElectroOptical Innovations
55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058
hobbs at electrooptical dot net
http://electrooptical.net