Clock Frequency Synthesis

I suggest that you ditch the idea of doing it with a switched capacitor filters and frequency synthesis, and instead do it by direct digital synthesis and some ADCs. One reasonably fast microprocessor (32-bit ARM Cortex) should be able to handle six channels with ease.

--
Tim Wescott
Control system and signal processing consulting
www.wescottdesign.com

Presumably this means spacing them by about 180Hz - 100Hz, 280Hz ...

820Hz, 1kHz.

If you started off with a 64kHz oscillator, you'd divide that by 640 to get 100Hz, 256 to get 250Hz, by 128 to get 500Hz, by 100 to get

640Hz, by 80 to get 800Hz, and by 64 to get 1kHz - which would all fit inside a single programmable logic device.

If you want eight 167 times faster clocks for the switched capacitor filters, you could multiply 64kHz by 167 giving you a clock frequency of 10.688MHz, which is pretty much the ideal crystal frequency, and complicate your programmable logic device appropriately.

With a higher starting frequency you could get more evenly spaced modulation frequencies that were exact multiples of 1Hz. The programmable logic device would start drawing lots of current if you did.

-- Bill Sloman, Nijmegen

f . M

e quoted text -

I've only used a switched cap filters once. LTC1063. There were (what we called) intermodulation distortion at the ~45-50dB level. It took a while to identify the SCF as the source.

Switched cap filters sound so nice 'in theory'.

To the spflanze; why so low frequencies? Wouldn't it be easier to work up at maybe the 10kHz range... and put 1kHz (or so) between channels? And then forget about the harmonics. Or one nice analog low pass after the photodiodes to kill them all.

George H.

Not enough to make it warm. Modern FPGAs and CPLDs can do a lot of logic at a 50 MHz clock for a few mA of Icc.

John

Yep. 50MHz is _audio_ with today's chips :-) ...Jim Thompson

--
| James E.Thompson, CTO                            |    mens     |
| Analog Innovations, Inc.                         |     et      |
| Analog/Mixed-Signal ASIC's and Discrete Systems  |    manus    |
| Phoenix, Arizona  85048    Skype: Contacts Only  |             |
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  |
| E-mail Icon at http://www.analog-innovations.com |    1962     |
             
I love to cook with wine.     Sometimes I even put it in the food.

Those would actually be DACs, but otherwise I completely agree. The DDS is simple, running 6 channels might use up a lot of I/O pins unless he uses serial DACs. A very simple filter at the output of the DAC will remove any quantization and other noises and result in a VERY pure signal.

Jon

This thing is supposed to be battery operated.

Actually, a more rational choice of frequencies is based on 10^1/6,

1.47

which gives you the series 100Hz, 147Hz, 215Hz, 317Hz, 466Hz, 681Hz and 1kHz which doesn't lend itself to dividing down from a single clock, since 147 and 317 are prime numbers.

The 6-bit SN7497 binary rate multiplier could cope with this sort of problem - it did generate an irregular pulse train, so you had to follow it with a regular divider that reduces the pulse rate by 2^6 to smooth out the variations.

2^6 is 64, so if we started with a 81.92kHz clock, set the binary rate multiplier to 50, then divide by 64, we'd get out 1kHz, while 34, would give us 680Hz, 23 would give us 460Hz, 16 would give us 320Hz, 11 would give us 220Hz and, 7 would give 140Hz and 5 would give us 100Hz - close enough for government work.

Getting the 167-fold higher frequencies for the switched capacitor filters means duplicating the logic from a 13.68064MHz clock, which is entirely practical.

You can obviously duplicate the internal logic of 7497 inside the programmable logic of your choice.

-- Bill Sloman, Nijmegen

-- Bill Sloman, Nijmegen

DDS with ADC's would be a challenge, wouldn't it?

Dyslexics Untie!!

--
My liberal friends think I'm a conservative kook.
My conservative friends think I'm a liberal kook.
Why am I not happy that they have found common ground?

Tim Wescott, Communications, Control, Circuits & Software
http://www.wescottdesign.com

instead of messing with a bunch of logic, use something like an stm32f4xx cortex ARM with lots of timers, capable of running at +100MHz

-Lasse

el

n e m t

he

of

en

f

s.

ARM

de quoted text -

The frequencies are chosen where I expect the noise in the system to be lowest. This is above the TIA's 1/f noise corner and beneath the zero created by the capacitance around the TIA's inverting input and feedback resistor. The frequencies of the six channels are not the same because I am expecting a little bit of the LED light to crosstalk between channels. All six optrode chemistries are on the same glass slide.

The six frequencies must be close together because they must all fit between the 1/f noise corner and the zero. If I use a divide by N counter, choose only those output frequencies that evenly divide into the reference frequency, have a large enough input reference frequency, and enough bits in the the counter, I will have enough frequency choices.

I am not using a DDS because of the quantization noise. If I use a DDS with a 10 bit DAC its quantization noise will dominate. It is white noise that will appear within the bandwidth of the FFT so there is no filtering it out. I would be using a 16bit AD7606 without the benefit of its 16bit quantization noise. The quantization noise calculation I have done shows fewer bits will cause quantization noise to dominate. I do not know a 16 bit DDS chip. Available 14 bit DDS chips consume too much power.

My opinion on whether the quantization noise output by a DDS is white or not has changed a few times. After looking at the figure 20.3b on pdf page 4 of:

(I have recently gotten a copy of this book have have yet to fully read it.) and also reading the conclusion at: sinusoidal/ I believe the white noise approximation holds. And I believe this is most especially true if the DDS tuning word is such that the phase the sine wave cycle starts at changes from cycle to cycle. This is most of the time. The exception being when the tuning word divides evenly into

2^N where N is the bit size of the tuning word register and adder.

I am also wondering if I should just square wave modulate the LED light and count on the FFT to extract the fundamental. So for the first version of this system I am going to include an analog switch that can bypass the filtering.

f

it

y

From a battery? At least with programmable logic, every switching event is one that you have programmed.

-- Bill Sloman, Nijmegen

nnel

an

rd

nse

to

am

s.

It

the

ut of

when

of

1 s.

or

t ARM

Hidequoted text -

t g

Your TIA and photodiode have a 3dB point near 1kHz? Big photodiode (PD)? Large feedback resistance? Are you reverse biasing the PD? It will pay you big time to make the front end faster.

More light (lower feedback R) PD no bigger than you need (lower C) reverse bias the PD (lower C) faster opamp. (max TIA freq goes as sqrt of opamp GBW)

Then for more exotic tricks you can bootstrap the PD, or try Phil Hobb's cascode on the front end. Both are attempts to reduce the effects of the PD capacitance.

George H.

quantized-sinusoidal/

So generate a signal from a micro, and feed it to a DAC with the precision of your choice. When I said it would be easy for the processor, I meant it would be easy _without_ using a dedicated DDS chip.

You can even dither the signal in such a way that the quantization noise turns blue and is thus easy to filter out.

Note that if there's cross-talk and if you really need 96dB of dynamic range (as indicated by your proposed use of 16-bit ADCs), you probably want to choose signal frequencies that not only fit, but whose harmonics don't fall on one another. You're almost certainly going to get distortion; you wouldn't want your 5th harmonic from one light landing right on the fundamental of another.

--
My liberal friends think I'm a conservative kook.
My conservative friends think I'm a liberal kook.
Why am I not happy that they have found common ground?

Tim Wescott, Communications, Control, Circuits & Software
http://www.wescottdesign.com

hannel

th an

hird

ponse

A to

I am

ips.

s. It

to the

tput of

n when

th of

or 1 s.

itor

l

bit ARM

m-Hidequotedtext -

=A0It

log

The feedback resistor is not known yet and will not be known until it is used. A digital potentiometer will be used in the feedback path because of this. So the location of the zero is not known now.

I want all six channels to be near the 1/f noise corner. For exceptional opamps this can be as low as 300Hz. For a typical op amp used for a TIA it could be as high as 1kHz.

I know about reducing PD capacitance. Besides the way Phil Hobbs does it, another way is the bootstrap with the JFET shown here:

Other bootstrap circuits are described in "Photodiode Amplifiers" by Jerald Graeme.

Photodiode capacitance can be reduced but not eliminated. And there are other capacitances besides the photodiode. The zero will exist no matter what. Wherever it is I want to stay under it.

of

..

40

fit

or

ncy

d

ou

o e s

If the timers on the processor can do the job it will greatly reduce the number of counter chips I have on the board. Its clock rate is much slower than the Blackfin but the FPU might make up for that. I am looking at it.

channel

.

with an

o

he

third

esponse

PGA to

cy I am

chips.

des. It

into the

output of

pen when

ngth of

s or 1 s.

acitor

tal

2-bit ARM

com-Hidequotedtext-

re

=A0It

o

nalog

k

ar.com/product/LTC6244

OK sounds like you're all set. I'll paraphrase one piece of wisdom from Phil H's book, (which would be worth your money), "Do the photon budget 'early'."

George H.

I'd be nice if that zero was out at

h

le of

...

640

et

ll fit

itor

uency

and

d

you

f f

to

ate

,

is

Use a single - big-enough - programmable logic device and bury all your counters (and the rest of the logic) inside it.

I've got a weakness for the Xilinx CoolRunner parts - they don't draw much current at all. I've never actually been able to use one - not for want of trying (I've got 15 of then in my desk drawer) - but they look good.

-- Bill Sloman, Nijmegen

f h y . n

(what

k a

ork

ter

k I f r

ttdesign.com

I intend the range of frequencies be much less than the 2nd harmonic of the low end of the range. Ideally the distance between each of them would be the minimum required which would be 1/T where T is the length of time the FFT is done over.

I have decided to abandon the switched capacitor filter and try your suggestion to use DACs that are constantly updated by the processor. The Blackfin's clock speed is 600MHz which I expect would be enough to keep the DAC's updated and do FFT in real time on 12 channels ( 6 signal plus 6 LED monitor ).

My current DAC choice:

products/product.html

I will update the DACs at the same rate as the ADC (200ksps). The DAC and ADC being correlated in time brings up the interesting question of what quantization noise will be contributed by the DAC. The correlated means the square root of the sum of the squares does not apply. Since the DAC's full scale will always be in use its effective resolution would a little bit greater than the ADC. One might consider the DAC won't add any quantization noise. And what would this mean for the benefit of having a reconstruction filter? Maybe none.

fit

--
"Not for want of trying"???

The parts languish, according to you, in your desk drawer, and you've
done nothing with them, and justify your sloth, how?