DDS Selection

I am designing designing a system that uses changes in the
transmissivity light in a detection chemistry to detect gas. The light
source is a sine wave modulated LED source. After the light passes
through the detection chemistry it will be measured by a photodiode
and a transimpedance amplifier circuit. The output of the
transimpedance amplifier is measured by a 16 bit AD7606 ADC which does
a DMA data transfer into a Blackfin DSP where a Fourier Transform is
done to extract an amplitude at the LED's modulation frequency.

There will be six detection channels. Each channel is to be modulated
at a different frequency to reduce crosstalk. The length of time the
Fourier Transform is done over will be in multiples of 100ms for
maximum rejection of both 50Hz and 60Hz. The modulation frequencies
are chosen in multiples of the inverse of this length of time for
maximum crosstalk rejection.

Between the output of the transimpedance amplifier and the ADC is an
anti-aliasing filter. Its cutoff frequency and number of polls are
chosen so the attenuation at the ADC's sampling frequency is equal or
less than the ADC's LSB divided by the ADC's total number of states
(2^16).

The system needs an option for battery power so current draw is an
issue.

I need to choose a DDS chip to generate the sine wave reference for
the LED's modulation. I have been looking at 14 Bit DDS chips at:
http://www.analog.com/ps/psthandler.aspx?pstid=10068&la=en
I notice there is a dramatic increase in power requirements the more
bits the DAC part of the DDS has. I would like to choose a 10 bit one
such as AD9838 that uses little power but I am concerned about the
quantization noise that appears as a spurs in the DDS output beginning
at twice the output frequency. That spur would be within the anti-
alias filter's passband. Theoretically this spur would appear right in
one of the notches of the sync function that is the frequency response
of a Fourier Transform and so would be taken out by it. In practice
can I actually count on it doing that? The DDS has the same frequency
reference as the ADC's sample rate so the spur would be precisely
located there.

Or do I need a reconstruction filter for the DDS that would have a cut
off between the modulation frequency and the first spur at twice this
frequency? I am sure it would need many poles. If so could the anti-
alias filter, although it is at the output of the transimpedance
amplifier instead of the DDS, double as a reconstruction filter if it
would have the same cutoff frequency and poles?

Or do I need a reconstruction filter for the DDS that would have a cut
off between the modulation frequency and the first spur at twice this
frequency? I am sure it would need many poles. If so could the anti-
alias filter, although it is at the output of the transimpedance
amplifier instead of the DDS, double as a reconstruction filter if it
would have the same cutoff frequency and poles?

I am designing designing a system that uses changes in the
transmissivity light in a detection chemistry to detect gas. The light
source is a sine wave modulated LED source. After the light passes
through the detection chemistry it will be measured by a photodiode
and a transimpedance amplifier circuit. The output of the
transimpedance amplifier is measured by a 16 bit AD7606 ADC which does
a DMA data transfer into a Blackfin DSP where a Fourier Transform is
done to extract an amplitude at the LED's modulation frequency.

There will be six detection channels. Each channel is to be modulated
at a different frequency to reduce crosstalk. The length of time the
Fourier Transform is done over will be in multiples of 100ms for
maximum rejection of both 50Hz and 60Hz. The modulation frequencies
are chosen in multiples of the inverse of this length of time for
maximum crosstalk rejection.

Between the output of the transimpedance amplifier and the ADC is an
anti-aliasing filter. Its cutoff frequency and number of polls are
chosen so the attenuation at the ADC's sampling frequency is equal or
less than the ADC's LSB divided by the ADC's total number of states
(2^16).

The system needs an option for battery power so current draw is an
issue.

I need to choose a DDS chip to generate the sine wave reference for
the LED's modulation. I have been looking at 14 Bit DDS chips at:
http://www.analog.com/ps/psthandler.aspx?pstid=10068&la=en
I notice there is a dramatic increase in power requirements the more
bits the DAC part of the DDS has. I would like to choose a 10 bit one
such as AD9838 that uses little power but I am concerned about the
quantization noise that appears as a spurs in the DDS output beginning
at twice the output frequency. That spur would be within the anti-
alias filter's passband. Theoretically this spur would appear right in
one of the notches of the sync function that is the frequency response
of a Fourier Transform and so would be taken out by it. In practice
can I actually count on it doing that? The DDS has the same frequency
reference as the ADC's sample rate so the spur would be precisely
located there.

Or do I need a reconstruction filter for the DDS that would have a cut
off between the modulation frequency and the first spur at twice this
frequency? I am sure it would need many poles. If so could the anti-
alias filter, although it is at the output of the transimpedance
amplifier instead of the DDS, double as a reconstruction filter if it
would have the same cutoff frequency and poles?

I am designing designing a system that uses changes in the
transmissivity light in a detection chemistry to detect gas.

spflanze wrote:
I am designing designing a system that uses changes in the
transmissivity light in a detection chemistry to detect gas.

You designing a homework.

Or do I need a reconstruction filter for the DDS that would have a cut
off between the modulation frequency and the first spur at twice this
frequency? I am sure it would need many poles. If so could the anti-
alias filter, although it is at the output of the transimpedance
amplifier instead of the DDS, double as a reconstruction filter if it
would have the same cutoff frequency and poles?

If we did a mental exercise and reduced the bits in the DDS, eventually
it would be a square wave. So wouldn't the first noise appear at 3x the
modulation frequency if the modulation frequency and clock rate have an
integer relationship. If so, then your post filter is substantially
easier to build.

spflanze wrote:
I am designing designing a system that uses changes in the
transmissivity light in a detection chemistry to detect gas. The light
source is a sine wave modulated LED source. After the light passes
through the detection chemistry it will be measured by a photodiode
and a transimpedance amplifier circuit. The output of the
transimpedance amplifier is measured by a 16 bit AD7606 ADC which does
a DMA data transfer into a Blackfin DSP where a Fourier Transform is
done to extract an amplitude at the LED's modulation frequency.

There will be six detection channels. Each channel is to be modulated
at a different frequency to reduce crosstalk. The length of time the
Fourier Transform is done over will be in multiples of 100ms for
maximum rejection of both 50Hz and 60Hz. The modulation frequencies
are chosen in multiples of the inverse of this length of time for
maximum crosstalk rejection.

Between the output of the transimpedance amplifier and the ADC is an
anti-aliasing filter. Its cutoff frequency and number of polls are
chosen so the attenuation at the ADC's sampling frequency is equal or
less than the ADC's LSB divided by the ADC's total number of states
(2^16).

The system needs an option for battery power so current draw is an
issue.

I need to choose a DDS chip to generate the sine wave reference for
the LED's modulation. I have been looking at 14 Bit DDS chips at:
http://www.analog.com/ps/psthandler.aspx?pstid=10068&la=en
I notice there is a dramatic increase in power requirements the more
bits the DAC part of the DDS has. I would like to choose a 10 bit one
such as AD9838 that uses little power but I am concerned about the
quantization noise that appears as a spurs in the DDS output beginning
at twice the output frequency. That spur would be within the anti-
alias filter's passband. Theoretically this spur would appear right in
one of the notches of the sync function that is the frequency response
of a Fourier Transform and so would be taken out by it. In practice
can I actually count on it doing that? The DDS has the same frequency
reference as the ADC's sample rate so the spur would be precisely
located there.

Or do I need a reconstruction filter for the DDS that would have a cut
off between the modulation frequency and the first spur at twice this
frequency? I am sure it would need many poles. If so could the anti-
alias filter, although it is at the output of the transimpedance
amplifier instead of the DDS, double as a reconstruction filter if it
would have the same cutoff frequency and poles?

All that just to modulate a LED with sine wave output? Sounds a bit
over kill to me.

One item on the list you need to consider, the LED is not linear in
its transition to current verses emissions. The first item on the list
should be a driving circuit using a photo feed back to ensure you have a
linear representation of photons emitting through your chemistry.

Generating the SINE wave in short, can be done with out the use of DDS
technology, which just adds to the pile of over kill.

Using an analog method of generating that sine wave with a
synchronous signal for the receiver would be more to what I would expect
to use.

Jamie

Jamie wrote:

spflanze wrote:
I am designing designing a system that uses changes in the
transmissivity light in a detection chemistry to detect gas. The light
source is a sine wave modulated LED source. After the light passes
through the detection chemistry it will be measured by a photodiode
and a transimpedance amplifier circuit. The output of the
transimpedance amplifier is measured by a 16 bit AD7606 ADC which does
a DMA data transfer into a Blackfin DSP where a Fourier Transform is
done to extract an amplitude at the LED's modulation frequency.

There will be six detection channels. Each channel is to be modulated
at a different frequency to reduce crosstalk. The length of time the
Fourier Transform is done over will be in multiples of 100ms for
maximum rejection of both 50Hz and 60Hz. The modulation frequencies
are chosen in multiples of the inverse of this length of time for
maximum crosstalk rejection.

Between the output of the transimpedance amplifier and the ADC is an
anti-aliasing filter. Its cutoff frequency and number of polls are
chosen so the attenuation at the ADC's sampling frequency is equal or
less than the ADC's LSB divided by the ADC's total number of states
(2^16).

The system needs an option for battery power so current draw is an
issue.

I need to choose a DDS chip to generate the sine wave reference for
the LED's modulation. I have been looking at 14 Bit DDS chips at:
http://www.analog.com/ps/psthandler.aspx?pstid=10068&la=en
I notice there is a dramatic increase in power requirements the more
bits the DAC part of the DDS has. I would like to choose a 10 bit one
such as AD9838 that uses little power but I am concerned about the
quantization noise that appears as a spurs in the DDS output beginning
at twice the output frequency. That spur would be within the anti-
alias filter's passband. Theoretically this spur would appear right in
one of the notches of the sync function that is the frequency response
of a Fourier Transform and so would be taken out by it. In practice
can I actually count on it doing that? The DDS has the same frequency
reference as the ADC's sample rate so the spur would be precisely
located there.

Or do I need a reconstruction filter for the DDS that would have a cut
off between the modulation frequency and the first spur at twice this
frequency? I am sure it would need many poles. If so could the anti-
alias filter, although it is at the output of the transimpedance
amplifier instead of the DDS, double as a reconstruction filter if it
would have the same cutoff frequency and poles?

All that just to modulate a LED with sine wave output? Sounds a bit
over kill to me.

One item on the list you need to consider, the LED is not linear in
its transition to current verses emissions. The first item on the list
should be a driving circuit using a photo feed back to ensure you have a
linear representation of photons emitting through your chemistry.

Generating the SINE wave in short, can be done with out the use of DDS
technology, which just adds to the pile of over kill.

Using an analog method of generating that sine wave with a
synchronous signal for the receiver would be more to what I would expect
to use.

Jamie


Nah, LEDs are pretty linear above about 10 uA. I second the suggestion
to use square wave drive and a lock-in per channel, i.e. an AM radio
with synchronous detection. (AM radios are pretty good at adjacent
channel rejection, for such simple devices.) For instance, the OP can
pick drive frequencies of, say, 12, 13, 14, 15, 16, and 17 kHz, with 100
Hz-ish lowpass filters after the demodulator. One wants the low order
harmonics of each modulation frequency to miss each other, because that
helps a lot with the spurious signal reduction. If he doesn't want to
lose all that nice software work, he can write a program to find the
optimum channel assignments that minimize crosstalk.

Square wave lock-ins are sensitive to odd harmonics, but a reasonably
good lowpass filter after the TIA will get rid of those, as well as
noise contributions from their neighbourhoods.

Cheers

Phil Hobbs

On 1/22/2012 1:09 PM, Phil Hobbs wrote:
Jamie wrote:

spflanze wrote:
I am designing designing a system that uses changes in the
transmissivity light in a detection chemistry to detect gas. The light
source is a sine wave modulated LED source. After the light passes
through the detection chemistry it will be measured by a photodiode
and a transimpedance amplifier circuit. The output of the
transimpedance amplifier is measured by a 16 bit AD7606 ADC which does
a DMA data transfer into a Blackfin DSP where a Fourier Transform is
done to extract an amplitude at the LED's modulation frequency.

There will be six detection channels. Each channel is to be modulated
at a different frequency to reduce crosstalk. The length of time the
Fourier Transform is done over will be in multiples of 100ms for
maximum rejection of both 50Hz and 60Hz. The modulation frequencies
are chosen in multiples of the inverse of this length of time for
maximum crosstalk rejection.

Between the output of the transimpedance amplifier and the ADC is an
anti-aliasing filter. Its cutoff frequency and number of polls are
chosen so the attenuation at the ADC's sampling frequency is equal or
less than the ADC's LSB divided by the ADC's total number of states
(2^16).

The system needs an option for battery power so current draw is an
issue.

I need to choose a DDS chip to generate the sine wave reference for
the LED's modulation. I have been looking at 14 Bit DDS chips at:
http://www.analog.com/ps/psthandler.aspx?pstid=10068&la=en
I notice there is a dramatic increase in power requirements the more
bits the DAC part of the DDS has. I would like to choose a 10 bit one
such as AD9838 that uses little power but I am concerned about the
quantization noise that appears as a spurs in the DDS output beginning
at twice the output frequency. That spur would be within the anti-
alias filter's passband. Theoretically this spur would appear right in
one of the notches of the sync function that is the frequency response
of a Fourier Transform and so would be taken out by it. In practice
can I actually count on it doing that? The DDS has the same frequency
reference as the ADC's sample rate so the spur would be precisely
located there.

Or do I need a reconstruction filter for the DDS that would have a cut
off between the modulation frequency and the first spur at twice this
frequency? I am sure it would need many poles. If so could the anti-
alias filter, although it is at the output of the transimpedance
amplifier instead of the DDS, double as a reconstruction filter if it
would have the same cutoff frequency and poles?

All that just to modulate a LED with sine wave output? Sounds a bit
over kill to me.

One item on the list you need to consider, the LED is not linear in
its transition to current verses emissions. The first item on the list
should be a driving circuit using a photo feed back to ensure you have a
linear representation of photons emitting through your chemistry.

Generating the SINE wave in short, can be done with out the use of DDS
technology, which just adds to the pile of over kill.

Using an analog method of generating that sine wave with a
synchronous signal for the receiver would be more to what I would expect
to use.

Jamie


Nah, LEDs are pretty linear above about 10 uA. I second the suggestion
to use square wave drive and a lock-in per channel, i.e. an AM radio
with synchronous detection. (AM radios are pretty good at adjacent
channel rejection, for such simple devices.) For instance, the OP can
pick drive frequencies of, say, 12, 13, 14, 15, 16, and 17 kHz, with 100
Hz-ish lowpass filters after the demodulator. One wants the low order
harmonics of each modulation frequency to miss each other, because that
helps a lot with the spurious signal reduction. If he doesn't want to
lose all that nice software work, he can write a program to find the
optimum channel assignments that minimize crosstalk.

Square wave lock-ins are sensitive to odd harmonics, but a reasonably
good lowpass filter after the TIA will get rid of those, as well as
noise contributions from their neighbourhoods.

Cheers

Phil Hobbs


Why do a simple analog solution when you can throw a computer at the
problem. ;-)

I'm probably not following the need for a lock-in since he has all the
oscillator sources at hand. You don't have to lock, just demodulate the
photo signal using the modulation source. Quite possible I'm missing
something here.

On Sun, 22 Jan 2012 17:14:24 -0800, miso <m...@sushi.com> wrote:
On 1/22/2012 1:09 PM, Phil Hobbs wrote:
Jamie wrote:

spflanze wrote:
I am designing designing a system that uses changes in the
transmissivity light in a detection chemistry to detect gas. The light
source is a sine wave modulated LED source. After the light passes
through the detection chemistry it will be measured by a photodiode
and a transimpedance amplifier circuit. The output of the
transimpedance amplifier is measured by a 16 bit AD7606 ADC which does
a DMA data transfer into a Blackfin DSP where a Fourier Transform is
done to extract an amplitude at the LED's modulation frequency.

There will be six detection channels. Each channel is to be modulated
at a different frequency to reduce crosstalk. The length of time the
Fourier Transform is done over will be in multiples of 100ms for
maximum rejection of both 50Hz and 60Hz. The modulation frequencies
are chosen in multiples of the inverse of this length of time for
maximum crosstalk rejection.

Between the output of the transimpedance amplifier and the ADC is an
anti-aliasing filter. Its cutoff frequency and number of polls are
chosen so the attenuation at the ADC's sampling frequency is equal or
less than the ADC's LSB divided by the ADC's total number of states
(2^16).

The system needs an option for battery power so current draw is an
issue.

I need to choose a DDS chip to generate the sine wave reference for
the LED's modulation. I have been looking at 14 Bit DDS chips at:
http://www.analog.com/ps/psthandler.aspx?pstid=10068&la=en
I notice there is a dramatic increase in power requirements the more
bits the DAC part of the DDS has. I would like to choose a 10 bit one
such as AD9838 that uses little power but I am concerned about the
quantization noise that appears as a spurs in the DDS output beginning
at twice the output frequency. That spur would be within the anti-
alias filter's passband. Theoretically this spur would appear right in
one of the notches of the sync function that is the frequency response
of a Fourier Transform and so would be taken out by it. In practice
can I actually count on it doing that? The DDS has the same frequency
reference as the ADC's sample rate so the spur would be precisely
located there.

Or do I need a reconstruction filter for the DDS that would have a cut
off between the modulation frequency and the first spur at twice this
frequency? I am sure it would need many poles. If so could the anti-
alias filter, although it is at the output of the transimpedance
amplifier instead of the DDS, double as a reconstruction filter if it
would have the same cutoff frequency and poles?

     All that just to modulate a LED with sine wave output?  Sounds a bit
over kill to me.

     One item on the list you need to consider, the LED is not linear in
its transition to current verses emissions. The first item on the list
should be a driving circuit using a photo feed back to ensure you have a
   linear representation of photons emitting through your chemistry.

    Generating the SINE wave in short, can be done with out the use of DDS
   technology, which just adds to the pile of over kill.

     Using an analog method of generating that sine wave with a
synchronous signal for the receiver would be more to what I would expect
to use.

   Jamie

Nah, LEDs are pretty linear above about 10 uA.  I second the suggestion
to use square wave drive and a lock-in per channel, i.e. an AM radio
with synchronous detection.  (AM radios are pretty good at adjacent
channel rejection, for such simple devices.)  For instance, the OP can
pick drive frequencies of, say, 12, 13, 14, 15, 16, and 17 kHz, with 100
Hz-ish lowpass filters after the demodulator.   One wants the low order
harmonics of each modulation frequency to miss each other, because that
helps a lot with the spurious signal reduction.  If he doesn't want to
lose all that nice software work, he can write a program to find the
optimum channel assignments that minimize crosstalk.

Square wave lock-ins are sensitive to odd harmonics, but a reasonably
good lowpass filter after the TIA will get rid of those, as well as
noise contributions from their neighbourhoods.

Cheers

Phil Hobbs

Why do a simple analog solution when you can throw a computer at the
problem. ;-)

I'm probably not following the need for a lock-in since he has all the
oscillator sources at hand. You don't have to lock, just demodulate the
photo signal using the modulation source. Quite possible I'm missing
something here.

"Lock-in amplifier" refers to synchronous demodulation followed by
low-pass filtering.

http://en.wikipedia.org/wiki/Lock-in_amplifier

"Lock-in amplifier" refers to synchronous demodulation followed by
low-pass filtering.

http://en.wikipedia.org/wiki/Lock-in_amplifier

Strictly speaking, you've got to have a modulation on the signal being
detected before you can lock-in onto that modulation and synchronously
detect it.

Normally you excite the system under test with some sort of
alternating signal - sine wave are often handy, square wave are easier
to make, but repeated narrow pulses at regular intervals can be more
appropriate for some problems.

--
Bill Sloman, Nijmegen

http://www.boselec.com/products/siglimwhat.html

On 1/22/2012 1:09 PM, Phil Hobbs wrote:
Jamie wrote:

spflanze wrote:
I am designing designing a system that uses changes in the
transmissivity light in a detection chemistry to detect gas. The light
source is a sine wave modulated LED source. After the light passes
through the detection chemistry it will be measured by a photodiode
and a transimpedance amplifier circuit. The output of the
transimpedance amplifier is measured by a 16 bit AD7606 ADC which does
a DMA data transfer into a Blackfin DSP where a Fourier Transform is
done to extract an amplitude at the LED's modulation frequency.

There will be six detection channels. Each channel is to be modulated
at a different frequency to reduce crosstalk. The length of time the
Fourier Transform is done over will be in multiples of 100ms for
maximum rejection of both 50Hz and 60Hz. The modulation frequencies
are chosen in multiples of the inverse of this length of time for
maximum crosstalk rejection.

Between the output of the transimpedance amplifier and the ADC is an
anti-aliasing filter. Its cutoff frequency and number of polls are
chosen so the attenuation at the ADC's sampling frequency is equal or
less than the ADC's LSB divided by the ADC's total number of states
(2^16).

The system needs an option for battery power so current draw is an
issue.

I need to choose a DDS chip to generate the sine wave reference for
the LED's modulation. I have been looking at 14 Bit DDS chips at:
http://www.analog.com/ps/psthandler.aspx?pstid=10068&la=en
I notice there is a dramatic increase in power requirements the more
bits the DAC part of the DDS has. I would like to choose a 10 bit one
such as AD9838 that uses little power but I am concerned about the
quantization noise that appears as a spurs in the DDS output beginning
at twice the output frequency. That spur would be within the anti-
alias filter's passband. Theoretically this spur would appear right in
one of the notches of the sync function that is the frequency response
of a Fourier Transform and so would be taken out by it. In practice
can I actually count on it doing that? The DDS has the same frequency
reference as the ADC's sample rate so the spur would be precisely
located there.

Or do I need a reconstruction filter for the DDS that would have a cut
off between the modulation frequency and the first spur at twice this
frequency? I am sure it would need many poles. If so could the anti-
alias filter, although it is at the output of the transimpedance
amplifier instead of the DDS, double as a reconstruction filter if it
would have the same cutoff frequency and poles?

All that just to modulate a LED with sine wave output? Sounds a bit
over kill to me.

One item on the list you need to consider, the LED is not linear in
its transition to current verses emissions. The first item on the list
should be a driving circuit using a photo feed back to ensure you have a
linear representation of photons emitting through your chemistry.

Generating the SINE wave in short, can be done with out the use of DDS
technology, which just adds to the pile of over kill.

Using an analog method of generating that sine wave with a
synchronous signal for the receiver would be more to what I would expect
to use.

Jamie


Nah, LEDs are pretty linear above about 10 uA. I second the suggestion
to use square wave drive and a lock-in per channel, i.e. an AM radio
with synchronous detection. (AM radios are pretty good at adjacent
channel rejection, for such simple devices.) For instance, the OP can
pick drive frequencies of, say, 12, 13, 14, 15, 16, and 17 kHz, with 100
Hz-ish lowpass filters after the demodulator. One wants the low order
harmonics of each modulation frequency to miss each other, because that
helps a lot with the spurious signal reduction. If he doesn't want to
lose all that nice software work, he can write a program to find the
optimum channel assignments that minimize crosstalk.

Square wave lock-ins are sensitive to odd harmonics, but a reasonably
good lowpass filter after the TIA will get rid of those, as well as
noise contributions from their neighbourhoods.

Cheers

Phil Hobbs


Why do a simple analog solution when you can throw a computer at the
problem. ;-)

I'm probably not following the need for a lock-in since he has all the
oscillator sources at hand. You don't have to lock, just demodulate the
photo signal using the modulation source. Quite possible I'm missing
something here.

If the modulation circuit also produced a quadrature clock source, then
the photo signal could be IQ demodulated. One path would be the signal
level, and I suppose the other path would be a figure of merit of the
noise. [All assuming no significant phase shift in the photo signal.]

Lock-ins don't phase-lock to the input--otherwise you couldn't measure
the static phase. You supply them a reference, which in this simple
situation is the same as the drive signal, phase shifted to match the
pre-detection filter (e.g. the RF filter in an AM radio).
Alternatively, since it's only the amplitude he cares about, he could
phase-lock to the detected signal. That can get exciting in a
multichannel setup, though.

Cheers

Phil Hobbs


But you don't really care about the static phase shift, other than if it