DDS Selection

a bit

in

DDS

An analog feedback control circuit for driving the LED's using feedback from a monitoring photodiode is being considered. This is not just for linearity, but to keep the light output constant as the LED ages and also for noise reduction. The light output of an LED for the same current declines with age. I suspect also there is noise other than shot noise and noise in the driven current. I suspect something similar to 1/F noise in opamps from the LED alone that such a current control circuit would counter. I have never seen such a noise specification from an LED manufacturer. Attempts to get this from one went nowhere. The LED manufacturers make them for lighting and indicators. They neglect those of us who use them for optrodes.

I left the current control out of my initial description because it was not essential to the concern I started this thread about.

I was a bit careless in that first sentence. But me design no homework. Me design for company chemical warfare agent detector. :)

s a bit

r in

of DDS

d text -

Unless you really over drive them LEDs are pretty much just shot noise.. at least the few I've looked at. You have to drive them from a low noise source. (of course)

George H.

ounds a bit

inear in

y.

e of DDS

ct

on

er

We are looking for the smallest possible changes in transmissivity. So I am looking to do state of the art in noise performance.

Modulation is necessary to take out offset error drift in the electronics. Drift is really the lower end of the 1/F noise spectrum. I will choose a modulation frequency right where the 1/F noise spectrum gets buried in other noise. beyond this there is a point where noise starts increasing due to the zero made with capacitance around the TIA's virtual ground and the TIA's feedback resistor. I am doing sinewave modulation to keep the carrier frequency components beneath that zero. It is simpler if there is only one frequency component.

If both the modulating and demodulating waveforms are square wave there will be contributions to the output at odd harmonics where the signal to noise ratios of the system is less. The square wave is the equivalent of doing both a carrier signal and Fourier Transform at the fundamental and again at all odd harmonics and summing the results.

The odd harmonics can be filtered. A filter with enough attenuation at the first of the harmonics and very little attenuation at the carrier frequency would require many poles, and would require many op amps.

A sine wave can be generated by op amp circuits. But because for noise reduction I am looking to extract the carrier in as narrow band a filter as possible it means extracting it with an FFT. So the analog circuit generator has to be precise in frequency. That means phase locking its output which requires many parts and board area. A DDS can do it all in one chip.

The response time of this system is crucial. The box car averaging in the FFT offers the best compromises between sensitivity (determined by noise) and response time. The more time the FFT is done over the narrower its bandwidth.

The decision I need to make is whether I need 10, 12, or 14 bit DDS.

Not all the LEDs will be the same wavelength. Each will illuminate a different chemistry. For those LEDs of the same wavelength it is too difficult to split the beam to where the light needs to go.

Also it is known there will be cross talk. How much isn't known yet.

s a bit

r in

of DDS

LED's are photodiodes - if not very good one's - which may complicate life a bit

If the LED output is decaying during it's operational life, then the decay mechanism (whatever it is) is going to show up as a sort of 1/f noise. If the light output that you monitor doesn't come from exactly the saem part of the LED as the light output you are using to measure transmission, the feedback control of brightness may not cancel out this kind of 1/f noise. The decay mechanism could be localised in the position or the direction of the element of the light source that's changing, so you might not be controlling exactly what you want to control.

Because you don't form a significant portion of the market. You might do better if you bought laser diodes, but they are a lot more expensive.

-- Bill Sloman, Nijmegen

A careless designer building a chemical warfare agent detector. That's comforting.

A triangular wave is just the integral of a square wave, and has the same - odd only - harmonics, but with the amplitude dropoing in proportion to the square of the harmonic number.

But yes, the output from a DDS is potentially messier than the output from simpler switching circuit.

Have you thought about a delay-line filter? You clock your square wave through a shift register, hook up a suitable resistor to each stage, and sum the outputs. IIRR a sinc function tapered with a Hamming window to kills the Gibbs oscillation can give you a very clean sine wave; you've got residual high-frequency harmonics, but the resistors can be 0.1% parts (with a bit of padding to get the exact values) and the lower harmonics can be held to better than 60dB down.

-- Bill Sloman, Nijmegen

Well, you have to live with the results, so do what you're comfortable with. The lock-in will blow the doors off the digital approach for noise.

Cheers

Phil Hobbs

What is your target on transmissivity change and available light power ?

Can you take part of the light before the chemistry and use that as the reference for the intensity ? This with lock-in detection (done in HW or SW) has given good results for us in similar measurements.

e quoted text -

Cool, sounds like a fun project. One thing that stinks about LED's is you don't get a lot of light out of them.

George H.

Use a notch (or several as needed) as well as the LPF to get rid of the harmonics.

guess most of devices like that are mostly for security theater anyways, so it only has to work enough to convince those who write the checks in DHS/TSA

-Lasse

text -

Don't know for noise spec but I once encountered a low frequency issue which was traced down to convection combining with the more often spec'd LED output efficiency tempco. (the LEDs were driven withing their rated specs)

So how 'good' can you do Mikko? I've looked at these really tiny changes in our optical pumping apparatus, (a straight DC light level measurement, no modulation or lockin) and seen changes at smaller than a part in 10^5. (That was averaging several sweeps though.. Signal to noise in one sweep is about 1 at changes of a part in 10^5.)

George H.

cut

is

it

ly

he

an

It is not a complex thing for the ADC and the DDS to have the same external clock reference. If both I and Q are required because there is enough phase shift in the electronics, then implementing it is just a matter of software.

Notches aren't a good idea inside a lock-in amplifier set-up. Too much phase shift.

You start needing to add all-pass networks to your filter networks to keep the phase-shift low and stable.

-- Bill Sloman, Nijmegen

The photodiode would look at the LED sideways, which would be outside its lens's cone, but would still have plenty of light due to its proximity. After Bill Sloman's response I am not sure about noise suppression. I would expect a photodiode in such an arrangement to get more light from some areas of the light emitting surface than others. I have some lab experience where the feedback seemed to suppress noise on the output of a Luxeon III. I do not know where the monitoring photodiode was relative to this LED because both were in a sealed unit. The two LEDs being considered now are:

There is not yet a specific target.

DDS

Why not use square wave modulation and a receive TIA with fairly wide bandwidth, take lots of ADC samples, and do the lockin thing digitally? That will dither the ADC codes into oblivion. TIA phase shift will be small enough to ignore.

John