Update on laser distance meter project (master thesis) and question

[Yannick]

Maybe some remember i was developing a laser range finder for my
master thesis. Now 2 months from the end i can say it is a succes. I
can measure distances up to 10meters from white objects. The
resolution is good the first few meters (1-2cm) and afterwards worse
and worse (10-15cm for 10 meter). The problems lies on several things
, i still use a dds frequency generator with 125Mhz clock , for the
higher frequenties (20-30Mhz) the sinus is far from perfect. The
second problem is ofcourse the photodiode preamplifier wich is still
designed around the opa657 in the most simple configuration. The
thirth bottle neck is that i use a phase detector ic from analog
devices (AD8302) wich does clipping on the signal before multiplying.
Soo signals lower then noise cant be detected, this will be solved due
using a DSP processor (ad's blackfin) and with signal averaging i can
measure signals lower then noise (it will only take more time).

There is still something strange with the photodiode preamplifier , i
tried many things including the R-C-R trick described by winfield
hill, although simulation in microcap gave me stunning results and
huge bandwidth improvement, in practice it didnt work, i the best case
i had exact the same results and mostly worse. Then i thought of
something simpler, just using a resistor tree network to reduce
resistor values and still have the same total feedback resistor (68K
in my case). This is described in photodiode amplifiers from gerald
greame and i did it exactly like he pointed out. i use for R1 = 22K ,
R2=270 and R3 to mass = 100 soo i will have a total feedback
resistance of Rfeq=R1*(1+R2/R3) = 81.4K.

In simulation this worked wonderfull,more gain, more bandwidth then
with my singe 68K feedback resistance. But in practice the result was
worse. What can this be? everything is made in smd components with a
open ground plane to reduce capacitance to ground and everything is
shielded in a RFI case.

I want to thank Colin,Winfield hill and al the others for their help
in the beginning (6 months ago) of this project , it made everything a
lot easier for me.

greetings,

Yannick

 

[John Woodgate]

I read in sci.electronics.design that Yannick
<[email protected]>) about 'Update on
laser distance meter project (master thesis) and question', on Fri, 1
Apr 2005:

In simulation this worked wonderfull,more gain, more bandwidth then
with my singe 68K feedback resistance. But in practice the result was
worse. What can this be? everything is made in smd components with a
open ground plane to reduce capacitance to ground and everything is
shielded in a RFI case.

At what frequency? What is the open-loop bandwidth of the op-amp? What
closed-loop gain are you expecting?

There really should not be a large difference between simulation and
reality in a case like this. I think your simulation maybe includes some
'ideal' parts.

 

[Rene Tschaggelar]

Maybe some remember i was developing a laser range finder for my
master thesis. Now 2 months from the end i can say it is a succes. I
can measure distances up to 10meters from white objects. The
resolution is good the first few meters (1-2cm) and afterwards worse
and worse (10-15cm for 10 meter). The problems lies on several things
, i still use a dds frequency generator with 125Mhz clock , for the
higher frequenties (20-30Mhz) the sinus is far from perfect. The
second problem is ofcourse the photodiode preamplifier wich is still
designed around the opa657 in the most simple configuration. The
thirth bottle neck is that i use a phase detector ic from analog
devices (AD8302) wich does clipping on the signal before multiplying.
Soo signals lower then noise cant be detected, this will be solved due
using a DSP processor (ad's blackfin) and with signal averaging i can
measure signals lower then noise (it will only take more time).

The sine from the DDS requires some lowpass filterung too.

There is still something strange with the photodiode preamplifier , i
tried many things including the R-C-R trick described by winfield
hill, although simulation in microcap gave me stunning results and
huge bandwidth improvement, in practice it didnt work, i the best case
i had exact the same results and mostly worse. Then i thought of
something simpler, just using a resistor tree network to reduce
resistor values and still have the same total feedback resistor (68K
in my case). This is described in photodiode amplifiers from gerald
greame and i did it exactly like he pointed out. i use for R1 = 22K ,
R2=270 and R3 to mass = 100 soo i will have a total feedback
resistance of Rfeq=R1*(1+R2/R3) = 81.4K.

In simulation this worked wonderfull,more gain, more bandwidth then
with my singe 68K feedback resistance. But in practice the result was
worse. What can this be? everything is made in smd components with a
open ground plane to reduce capacitance to ground and everything is
shielded in a RFI case.

Highspeed amplifier configurations have low value feedback
resistors. How about going to 50 Ohms in the first stage and use
standard amplifiers, such as the MAR series from minicircuit.
They deliver 20dB with each stage. Or as limiting amplifier use
a MC100EL16D, an ECL line receiver, is does 20 dB or so too.


Rene

 

[Winfield Hill]

Rene Tschaggelar wrote...

i tried many things including the R-C-R trick described by winfield
hill, although simulation in microcap gave me stunning results and
huge bandwidth improvement, in practice it didnt work, i the best
case i had exact the same results and mostly worse. [snip]

Did you include your actual total input-node capacitance (detector,
cable and connectors, opamp, pcb, etc.) in the simulation?

The R-C-R trick simply corrects for the effect of high-value resistor
self-capacitance, in an attempt to make it act like a perfect resistor
without any self capacitance, or with a calculated small amount.

But you still have to meet the amplifier open-loop-gain criteria that
insures the summing node looks like a short, conveying all the signal
through the feedback resistor and not allowing it to go into the node
capacitance instead. A single opamp transamp looks like Zin = Rf/A,
where A = f_T/f. The opamp's f_T has to be high enough so Zin is less
than Xc-in = 1 / 2pi f Cin at your highest frequency. This means that
f_T > Rf Cin 2pi f^2. Check out the f^2 term! That can be a pretty
severe requirement at high frequencies, and it usually means working
hard to get Cin very low. Low Cin leads to very small detectors, no
cable, and a detector right at the preamp node - even in the same
package with the transamp IC, etc., for many manufacturer's parts.

Alternately you can work on the f_T aspect of the equation. This can
lead to a composite amplifier configuration, another subject I've
written about here several times. If you're forced that far into the
proverbial corner, you'll be seeking another solution.

I'm thinking of Phil Hobbs' common-base input-stage trick, simplifying
the opamp loop-gain problem. The opamp then works at the transistor's
collector, where Cin = Cob is low. Relatively high light levels are
required for dc light-signal applications, because high currents are
required to separately bias the input transistor to make its Zin low.
But you're working with ac light signals, so you could customize the
transamp to have higher gains for ac signals. It's worth considering.

 

[Yannick]

At what frequency? What is the open-loop bandwidth of the op-amp? What

closed-loop gain are you expecting?

frequenties between 20-30Mhz. Gain bandwidth product of the opa657 is
1600Mhz. i expect a transimpedance gain of 80K (98db) till 25-30Mhz
(simulation) with the resistor tee feedback network like i described.

There really should not be a large difference between simulation and
reality in a case like this. I think your simulation maybe includes some
'ideal' parts.

total input capacitance, every resistors parasitic capacitance,
compensation of the opamp, only parasitic inductance is not included
but this shouldn't be a problem.

 

[Yannick]

Did you include your actual total input-node capacitance (detector,

cable and connectors, opamp, pcb, etc.) in the simulation?

yes i did, the C30902E avalanche photodiode has 1.6p for +200V reverse
bias,the opa657 a total input capacitance of 5.3p and with the pcb
nodes i added 5p. This gives me a total input capacitance of +-12pf.
The length between the photodiode and opamp node is about 1.5cm. i
know i have to reduce this to reduce input capacitance but for now
this is not possible because i am still testing with different
configurations and i only have one avalanche photodiode wich was
expensive.

Is it possible that the nodes from photodiode to the opamp add
significantly more then 5p capacitance ? if this is the case i don't
have to look further ofcourse.

The R-C-R trick simply corrects for the effect of high-value resistor
self-capacitance, in an attempt to make it act like a perfect resistor
without any self capacitance, or with a calculated small amount.

i understand. the total feedback impedance will increase for higher
frequenties due current wich flow to mass with the RC network in
between soo this corrects the fall off due the parasitic capacitance
of the feedback resistor.

I'm thinking of Phil Hobbs' common-base input-stage trick, simplifying
the opamp loop-gain problem. The opamp then works at the transistor's
collector, where Cin = Cob is low. Relatively high light levels are
required for dc light-signal applications, because high currents are
required to separately bias the input transistor to make its Zin low.
But you're working with ac light signals, so you could customize the
transamp to have higher gains for ac signals. It's worth considering.

The common base aprouch is indeed a consideration i thought of but i
supposed my input capacitance was too low to have any effect but i
could be wrong. what kind of transistor should i use? i suppose a
tranistor with a very large transconductance Gm because Zin = 1/Gm for
common base, do you have any smd part numbers?.

I think i underestimated the total input capacitance, maybe this is
much larger then i supposed it was.

I also am cooling my avalanche photodiode with a peltier element to
increase its internal gain (responsitivity) and reduce shot noise ,
this works great but have to be carefull for condensation.


Yannick

 

[[email protected]]

Bipolar transistor transconductance is given by the Ebers-Moll
equation, and is essentially part independent.

I'd look at a low power wideband transistor - I've worked with the 5GHz
BFR92 (npn) and BFT92 (pnp) myself, but that was sometime ago. Farnell
now stocks some 25GHz Agilent parts - HBFP-0405 etc - and Infineon (was
Siemens) was trumpeing some 42GHz parts in their latest flyer.

 

[Winfield Hill]

Yannick wrote...

yes i did, the C30902E avalanche photodiode has 1.6p for +200V reverse
bias,the opa657 a total input capacitance of 5.3p and with the pcb
nodes i added 5p. This gives me a total input capacitance of +-12pf.
The length between the photodiode and opamp node is about 1.5cm.

OK, let's assume the total Cin = 12pF, Rf = 68k, corrected, with
a single opa657 opamp having f_T = 1600MHz, right?

fc = sqrt (f_T / Rf Cin 2pi) = 17.7MHz predicted bandwidth, plus
up to perhaps an extra 35% or so if your effective Rc is carefully
selected to take advantage of the two-pole response peaking effect.

Sorry, read it and weep. What are you seeking, just a little bit
faster, perhaps a 2x improvement. You're almost there but need a
little more? Let's see, you can place the PD right on the opamp,
with the connection wired in air, to reduce Cin a little, but that
might only improve the bandwidth by 20%. You could reduce your 68k,
and add back the lost gain later, and suffer a little more Johnson
noise (resistor Johnson noise may not be dominate anyway, see below).
You could increase the loop gain (composite configuration)...

Or you could consider another opamp with lower Cin. JFET's have to
be large for low en, but 5pF is a pretty high value in your scene.

BJT opamps have input-current shot noise, sqrt (2q I). Comparing
to resistor Johnson noise, sqrt (4kT/R), we get Ib < 50mV/Rf, so
we have to look for opamps whose Ib is less than 50mV/68k = 0.7uA.
But you also want a wide low-frequency extrapolated bandwidth, even
greater than 1600MHz (which means BJT opamps with input transistors
running at high currents), so finding the ideal part may be tricky.

BTW, what were your R-C-R values?

The common base aprouch is indeed a consideration i thought of but
i supposed my input capacitance was too low to have any effect but
i could be wrong.

I think that's true here.

what kind of transistor should i use? i suppose a tranistor with a
very large transconductance Gm because Zin = 1/Gm for common base,
do you have any smd part numbers?.

Any old small low-capacitance type will work. The transconductance
is always, g_m = Ie/Vt, so it's Ie that determines Zin = re.

For Cin = say 5pF total, you want re less than 530 ohms for a say
60MHz bandwidth, so Ie > Vt/re = 25mV/530 = 47uA. Call it 50uA,
which you can get from your +/-5V supplies. We see the BJT bias
resistors are larger than your chosen 68k feedback resistor, so
their deleterious impact on noise won't be too severe, or you can
get the bias current from say +/-15V to reduce the resistor noise.

.. +5
.. |
.. 100k ,- 68k ---,
.. | | __ |
.. ----+-- e c --+--+--|- \ |
.. | b | >--+---
.. 90k | ,--|+_/
.. | gnd |
.. -5V gnd

But as you say, this won't help much, because it's merely trading an
already-low APD capacitance for the transistor's Cob capacitance.

We haven't talked about w-en-Cin noise, which is always a killer at
high frequencies. en is 4.8nV for the opa657, which is pretty good,
but a selected low-noise BJT operating at 50uA might be a little bit
better. Also, many wideband bipolar opamps have less than 5nV noise.

 

[colin]

Maybe some remember i was developing a laser range finder for my
master thesis. Now 2 months from the end i can say it is a succes. I
can measure distances up to 10meters from white objects. The
resolution is good the first few meters (1-2cm) and afterwards worse
and worse (10-15cm for 10 meter). The problems lies on several things
, i still use a dds frequency generator with 125Mhz clock , for the
higher frequenties (20-30Mhz) the sinus is far from perfect. The
second problem is ofcourse the photodiode preamplifier wich is still
designed around the opa657 in the most simple configuration. The
thirth bottle neck is that i use a phase detector ic from analog
devices (AD8302) wich does clipping on the signal before multiplying.
Soo signals lower then noise cant be detected, this will be solved due
using a DSP processor (ad's blackfin) and with signal averaging i can
measure signals lower then noise (it will only take more time).

There is still something strange with the photodiode preamplifier , i
tried many things including the R-C-R trick described by winfield
hill, although simulation in microcap gave me stunning results and
huge bandwidth improvement, in practice it didnt work, i the best case
i had exact the same results and mostly worse. Then i thought of
something simpler, just using a resistor tree network to reduce
resistor values and still have the same total feedback resistor (68K
in my case). This is described in photodiode amplifiers from gerald
greame and i did it exactly like he pointed out. i use for R1 = 22K ,
R2=270 and R3 to mass = 100 soo i will have a total feedback
resistance of Rfeq=R1*(1+R2/R3) = 81.4K.

In simulation this worked wonderfull,more gain, more bandwidth then
with my singe 68K feedback resistance. But in practice the result was
worse. What can this be? everything is made in smd components with a
open ground plane to reduce capacitance to ground and everything is
shielded in a RFI case.

I want to thank Colin,Winfield hill and al the others for their help
in the beginning (6 months ago) of this project , it made everything a
lot easier for me.

greetings,

Yannick

Hi, I was wondering how you were getting on, and glad to of been some help
although i mentioned so many things i feel i might have confused things a
bit.

As we discued before ultimately the SNR for a given signal is limited by the
total capacitance at the detector and the amplifier input noise, regarldes
of any feedback network, wich can only degrade it further if you are not
carefull and might explain why its worse, or as before you experienced high
frequcncy instabliity wich you could not see on your scope.

there are several things to consider to make things better, starting at one
end and working through ..
1) stronger light source (probably not cheap or safe)
2) larger lense (cheap and easy)
3) APD cant remember if your using an avalnche detector or not (quite
expensive)
4) use a technique of downshifting the frequcncy using the photodiode itself
as a multiplier/mixer (for example 400khz), the result is that the signal is
no longer swamped by the efect of the capacitance and hence a great deal
stronger at the input to the amplifier, yet the actual phase shift in the
high frequcncy signal is maintained. you would need to generate a frequcncy
400khz lower or higher than your transmision signal and aply this to the
bias voltage of the detector, the folowing amplifier would only see signal
of 400khz so could easily be tuned further reducing the efect of the
capacitance. this may easily give a several thousandfold improvement in SNR,
its the technique i used in the end with dual transmision frequencies.
(not particularly easy but by far the most dramatic improvment and it
greatly simplifies the amplifier so may wel be worth it)

5) use a simple amplifier with a single ended input stage and much lower
noise than an op amp, such as one of the latest mosfets.

6) im not sure why you are focusing on constant bandwidth so much when you
are cliping the signal anyway, it seems like this is cuasing you a lot of
hardwork for no aparent gain.
what i would think is most important by far is its noise performance and
obviously that it has suficient gain and also phase delay that is
predictable and stable and that you can take this into acount in your phase
calculations.
reducing the feedback network to a bare minimum at the expense of flatness
of response might make things better - more stable and beter noise
performance. a constant 90' phase lag over the frequncy range you are using
might be quite easy to compensate for in your calculations, and extremly
easy to build a simple amplifier with much beter noise performance.

7) averaging over time is always posible, i asume you have tried taking
readings and averaging them by hand ? however i suspect that once the signal
drops below the noise floor at the clipper it is iretrevably lost.

8) im realy not sure to what extent the noise from the DDS is causing a
problem or how you would improve it, i suspect it has little efect, did i
sugest before a switched bank of sevaral crystal oscilators ? i have decided
to move away from PLL etc as its so hard to completly eliminate jitter from
creeping in through the supply ect, and use a series of dividers and
frequcncy multipliers to derive each frequency from one master clock, but i
only use 6 frequcnies in total.

Colin =^.^=

 

[colin]

I should also of mentioned to filter the signal carefully just before the
limiter, i think this is most important, if you have not done so already.
this should be quite easy. the efect is to eliminate noise outside of the
bandwidth you are using wich might be considerable.

COlin =^.^

 

[Yannick]

OK, let's assume the total Cin = 12pF, Rf = 68k, corrected, with

a single opa657 opamp having f_T = 1600MHz, right?

yes that's correct.

fc = sqrt (f_T / Rf Cin 2pi) = 17.7MHz predicted bandwidth, plus
up to perhaps an extra 35% or so if your effective Rc is carefully
selected to take advantage of the two-pole response peaking effect.
yes.

Sorry, read it and weep. What are you seeking, just a little bit
faster, perhaps a 2x improvement. You're almost there but need a
little more?

I want to improve the most whats possible in gain and bandwidth
without too much peaking because of the rapid phase change
(dphase/dfrequency) and with lowest noise possible. my goal is +-80k
transimpedance with a bandwidth of 30Mhz with a single amplifier. I
get this in simulation with using a feedback tee network (22K,270ohm
and 100ohm) and a OPA846 (bipolar).

Let's see, you can place the PD right on the opamp,

with the connection wired in air, to reduce Cin a little, but that
might only improve the bandwidth by 20%.

yes, i will do this at the end but for now i have to try to get the
max out of it with a little bit more Cin due the longer nodes.

You could reduce your 68k,

and add back the lost gain later, and suffer a little more Johnson
noise (resistor Johnson noise may not be dominate anyway, see below).

Yes but the problem is really the S/N wich is too low for the higher
frequenties. for my maximum distance of 10m i get the signals between
10 and 20Mhz quite good, but after 20Mhz rapid fall off soo at 30Mhz i
have a bad S/N ratio to have good resolution. For the closer distances
my 30Mhz signal is still +-100mv , this is good enough for 1 cm
accuracy but at 10m this is only 20-30mv.I don't think another
amplifier is needed because it will add noise and because of the
dynamid range of 60db of my phase detector (from 0dbm to -60dbm) i can
detect very low signals soo it's really up to increase the S/N ratio.

You could increase the loop gain (composite configuration)...

yes i only tried it in simulation and this gave good results but i
first want to try the more simple tee network in practice because i
have a feeling that the composite amplifier will give me also
problems(stability) wich i dont have with the simulation.

Or you could consider another opamp with lower Cin. JFET's have to
be large for low en, but 5pF is a pretty high value in your scene.

BJT opamps have input-current shot noise, sqrt (2q I). Comparing
to resistor Johnson noise, sqrt (4kT/R), we get Ib < 50mV/Rf, so
we have to look for opamps whose Ib is less than 50mV/68k = 0.7uA.
But you also want a wide low-frequency extrapolated bandwidth, even
greater than 1600MHz (which means BJT opamps with input transistors
running at high currents), so finding the ideal part may be tricky.

Yes i made a mathcad file with the signal noise calculations and the
OPA846 gave the best results for a feedback resistor of 68K and the
other parameters like i described them.

BTW, what were your R-C-R values?

Rf=80K,Cf=0.2p,R2=160,R3=53,C2=100pf


is always, g_m = Ie/Vt, so it's Ie that determines Zin = re.

Yes ofcourse

For Cin = say 5pF total, you want re less than 530 ohms for a say
60MHz bandwidth, so Ie > Vt/re = 25mV/530 = 47uA. Call it 50uA,
which you can get from your +/-5V supplies. We see the BJT bias
resistors are larger than your chosen 68k feedback resistor, so
their deleterious impact on noise won't be too severe, or you can
get the bias current from say +/-15V to reduce the resistor noise.

. +5
. |
. 100k ,- 68k ---,
. | | __ |
. ----+-- e c --+--+--|- \ |
. | b | >--+---
. 90k | ,--|+_/
. | gnd |
. -5V gnd

But as you say, this won't help much, because it's merely trading an
already-low APD capacitance for the transistor's Cob capacitance.

We haven't talked about w-en-Cin noise, which is always a killer at
high frequencies. en is 4.8nV for the opa657, which is pretty good,
but a selected low-noise BJT operating at 50uA might be a little bit
better. Also, many wideband bipolar opamps have less than 5nV noise.

Yes i understand the Eni noise , due the input capacitance this will
give noise peaking in the higher frequency region (till a fall off due
the GBP).i understand it like a 1+Zf/Zin amplification of the noise.
With the OPA657 this is indeed the dominant noise factor but with the
opa846 it's the bias current shot noise of 13pa/rootHz, the Eni there
is 1.3nv/rootHz.

I still cant figure out why the tee network described in greames book
doesnt work in practice for me.

 

[colin]

dont know if youve seen this ...

http://www.leica-geosystems.com/cpd/en/products/laser_distancemeter/lgs_4019.htm

i considered buying one to see whats inside it

Colin =^.^=

 

[Rene Tschaggelar]

7) averaging over time is always posible, i asume you have tried taking
readings and averaging them by hand ? however i suspect that once the signal
drops below the noise floor at the clipper it is iretrevably lost.

When the signal is kown, it can be recovered from far below the noise
by synchroneous rectification, physicists call it lock-in-amplifier.
AFAIK, the signal is known, it is the frequency sent out,
received with some less amplitude and an arbitrary delay/phase.

Rene

 

[Yannick]

Hi, I was wondering how you were getting on, and glad to of been some help
although i mentioned so many things i feel i might have confused things a
bit.

more because of my lack of knowledge, i think i understand it now all
much better then 6 months ago.

As we discued before ultimately the SNR for a given signal is limited by the
total capacitance at the detector and the amplifier input noise, regarldes
of any feedback network, wich can only degrade it further if you are not
carefull and might explain why its worse, or as before you experienced high
frequcncy instabliity wich you could not see on your scope.

there are several things to consider to make things better, starting at one
end and working through ..
1) stronger light source (probably not cheap or safe)

now 3mw, i dont want to go higher due safety reasons.

2) larger lense (cheap and easy)

is already quite large

3) APD cant remember if your using an avalnche detector or not (quite
expensive)

yes i use the same as you, C30902E at 200-220V reverse bias and cooled
with a peltier element.

4) use a technique of downshifting the frequcncy using the photodiode itself
as a multiplier/mixer (for example 400khz), the result is that the signal is
no longer swamped by the efect of the capacitance and hence a great deal
stronger at the input to the amplifier, yet the actual phase shift in the
high frequcncy signal is maintained. you would need to generate a frequcncy
400khz lower or higher than your transmision signal and aply this to the
bias voltage of the detector, the folowing amplifier would only see signal
of 400khz so could easily be tuned further reducing the efect of the
capacitance. this may easily give a several thousandfold improvement in SNR,
its the technique i used in the end with dual transmision frequencies.
(not particularly easy but by far the most dramatic improvment and it
greatly simplifies the amplifier so may wel be worth it)

you mentioned this 6 months ago but then i didnt understand it because
i didnt know much about mixing signals, now i think i understand.Did
you came up with it yourself or did you read it somewhere? But i think
it has to be very difficult because you have to sit in the exponential
part (but still in reverse bias) of the Id-Vak curve of the photodiode
otherwise it wont work. but i wonder if it still works for high
frequenties because the current wich flows due the light on the APD
will always see the input capacitance soo the 400Khz wich you get out
is indeed a lower frequency but will also be small if the main
frequency is very high (the carrier). The improvement will be on the
side of the feedback resistor wich can be much larger because the
bandwidth has only to be 400Khz of the transimpedance. Is this correct
or am i missing the point?

5) use a simple amplifier with a single ended input stage and much lower
noise then an op amp, such as one of the latest mosfets.

i might do this with the BF998 orso..

7) averaging over time is always posible, i asume you have tried taking
readings and averaging them by hand ? however i suspect that once the signal
drops below the noise floor at the clipper it is iretrevably lost.

I meant with signal averaging not just take phase averages , this is
what i do now in my microcontroller but i mean to detect the signal if
it is lower then noise.If your signal is periodic and you know the
frequency you can easily integrate it out of the noise , thats what i
meant with signal averaging. I am gonne do this later when i will
change my phase detector AD8302 with a fast ADC and a DSP (blackfin).
Now i cant due the clipping wich occurs inside the AD8302.

thanks for the good info again, are you still working on yours and
whats the resolution now and the maximum distance? I don't know if you
are cooling your acalanche photodiode but i think it also will give a
*2-3 improvement. The downside is the power consumption of a peltier
element soo battery applications is not possible.


Yannick

 

[LRCR]

i also did my Master Thesis (1999) on a Laser rangefinder where the
result was sampled, averaged and stored to an old C25 TI DSP.
Exciting project. Isn't it? I gather that you used a CW LASER. As
you discover, I found the photodiode/preamp can be quite tricky.

Caesar

 

[colin]

When the signal is kown, it can be recovered from far below the noise
by synchroneous rectification, physicists call it lock-in-amplifier.
AFAIK, the signal is known, it is the frequency sent out,
received with some less amplitude and an arbitrary delay/phase.

Rene

i think once a signal that is swamped with noise has been throught a limiter
or clipper your chances of recovering the signal are virtualy thrown away.
so much of the information is lost. its fine to do synchronous rectification
but once the signal has been cliped its to late, its realy so much beter to
do it before. if u look at a very noisy signal on an osciloscope that is
trigerd by the original signal you can make out the original quite easily,
onece its been cliped u cant.

 

[colin]

more because of my lack of knowledge, i think i understand it now all
much better then 6 months ago.

thats good

you mentioned this 6 months ago but then i didnt understand it because
i didnt know much about mixing signals, now i think i understand.

its actualy quite easy, much easier than you would expect, and as the
circuit around the photodiode is so simple theres little room for problems
to arise. the only complication is generating an extra frequency that is
always 400khz diferent from your transmited signal, this would be easy with
a second DSS programed 400khz apart. just inject it into the bias voltage
with a capacitor, however dont forget the bias voltage is 200v+ so u need a
high voltage capacitor, also when the bias voltage changes rapidly it can
feedback through the capacitor and blow up the driving device if your not
caefull.

Did
you came up with it yourself or did you read it somewhere?

i came up with it myself but cldnt beleive it hadnt been invented before as
with so many things i think of, however i had tried to find such a device
before but cldnt find one anywhere, but once i had designed this i managed
to find nasa had published a paper using a similar device. might dig out the
reference some time later but not got it to hand atm.

But i think
it has to be very difficult because you have to sit in the exponential
part (but still in reverse bias) of the Id-Vak curve of the photodiode
otherwise it wont work.

yes, but if you look at the slope of gain versus voltage it is very steep
especialy in the region of maximum realistic gain, but this is exactly where
you want to use it anyway, this means a smal variation in bias voltage ~
500mv wil cuase a very large change in gain, this is more than ample to
provide for a very efective multiplier.

the beauty is that if your signal is to strong you can reduce the bias
voltage and this reduces the gain of the photodiode and also the mixer gain,
making very efective agc. for smal change in bias voltage wich means minimal
change in apd capacitance.

but i wonder if it still works for high frequenties

yes it should scale up to high frequncis very well i am hoping to use this
technique to increasy frequcny up to 1ghz, limit would be lead
inducatance/device capacitance/transit time etc. but shld easily be at least
1ghz maybe much much more.

because the current wich flows due the light on the APD
will always see the input capacitance

wel yes and no, the mixing is done inside the device long before the high
frequcny sees the capacitance, in fact before it even gets converted to an
electrnic signal, its optoelectronic mixing, (note .. not optical mixing
wich is diferent and what i was trying to do before) in fact the capacitance
usefully serves to filter out the high frequency wich gets through wich is
no longer wanted, and also as a ground path for the mixing frequency.

is indeed a lower frequency but will also be small if the main
frequency is very high (the carrier).

no the mixing eficiency is very high and largly independant of the
frequcncy, it doesnt mater what the diference in frequcny is. the
capacitance stil has some considerable efect at 400khz, at lower frequcncies
the diference in signal strenght will be even higher however i found modern
electronic balast lamps played hell with it becuase the mixer does not
reject the lower light frequcncies.

for a 1:100 diference in frequecny you would see almost a 100 fold
improvment in signal strength and hence SNR, with modest tuning this will be
even higher say another 10 fold.

The improvement will be on the
side of the feedback resistor wich can be much larger because the
bandwidth has only to be 400Khz of the transimpedance. Is this correct
or am i missing the point?

you can forget TIA becuase you get a single frequcncy out of your mixer
stage. a simple voltage gain stage with no feedback resistor exept for dc
biasing is then by far the best. in fact the signal realy is so strong with
an apd runing at high multiplication you dont want too much amplification.

400khz bandwidth would be far to much, in fact the use of a narowband filter
reduces the noise and the reduction of bandwidth of 30mhz down to 3 khz
would give a further 100 fold reduction in noise.

this all adds up to give theoreticaly 100000 times improvement in SNR due to
the amplifier generated noise, theoreticaly matching the noise performance
of the APD.

i might do this with the BF998 orso..

its fairly easy to get it right with the simplest of circuits, it just
defers the problem of 90' phase lag acros the bandwidth to your final
calulations.

I meant with signal averaging not just take phase averages , this is
what i do now in my microcontroller but i mean to detect the signal if
it is lower then noise.If your signal is periodic and you know the
frequency you can easily integrate it out of the noise , thats what i
meant with signal averaging. I am gonne do this later when i will
change my phase detector AD8302 with a fast ADC and a DSP (blackfin).
Now i cant due the clipping wich occurs inside the AD8302.

ah ic yes if you change the phase detector to a linear one then you might
wel find a huge improvement anyway.

thanks for the good info again, are you still working on yours and
whats the resolution now and the maximum distance? I don't know if you
are cooling your acalanche photodiode but i think it also will give a
*2-3 improvement. The downside is the power consumption of a peltier
element soo battery applications is not possible.

I got mine to detect the smalest of movement of few mm or less over
considerable distance, however this was swamped by drift - it needs a
complete re doing of the prototype as its been cut up and changed so much,
the optics need to be fastend more securly, and higher grade npo capacitors
wich take up more board space than i had alowed, and an active filter as the
inductors i used for the low frequcncy norwoband filter cuase drift and
variation with signal strenght. I decided the FM ic i used to do so much of
the mixing and detecting is no longer apropriate.

I lost a bit of interest when i found they are comercialy available and
havnt done anything to it for some time, however my initial goal was to try
and resolve with much greater acuracy over more moderate distances for
machining aplications and so i need to increase the carier frequcny from
30mhz to upto 1ghz wich cuases more of a chalenge, the idea of using the
beat frequcny from 2 lasers is something i was considering, and wicked
several stage frquency multipler to get the mixing signal.

incidently i suspect that you are geting worse results when you improve the
high frequency bandwidth of you amplifier becuse your introducing a lot more
high frequency noise. a good steep filter before the detector might improve
things.

Colin.

 

[Rene Tschaggelar]

i think once a signal that is swamped with noise has been throught a limiter
or clipper your chances of recovering the signal are virtualy thrown away.
so much of the information is lost. its fine to do synchronous rectification
but once the signal has been cliped its to late, its realy so much beter to
do it before. if u look at a very noisy signal on an osciloscope that is
trigerd by the original signal you can make out the original quite easily,
onece its been cliped u cant.

A signal that is burried in the noise has unlikely being clipped.
The idea is to scan the phase of the reference through 2Pi and
look for the maximum and/or minimum. This together with a
synchroneous filter bandwidth of say 10Hz. Reducing the bandwidth
from 10MHz to 10Hz gives a gain of 10^6 or 60dB,(or perhaps 120dB ?)
This means you CAN get it out of the noise.
The angular phase resolution you can get out of the process is
questionable, especially when the time is limited.

Rene

 

[colin]

A signal that is burried in the noise has unlikely being clipped.
The idea is to scan the phase of the reference through 2Pi and
look for the maximum and/or minimum. This together with a
synchroneous filter bandwidth of say 10Hz. Reducing the bandwidth
from 10MHz to 10Hz gives a gain of 10^6 or 60dB,(or perhaps 120dB ?)
This means you CAN get it out of the noise.
The angular phase resolution you can get out of the process is
questionable, especially when the time is limited.

Well you can't rely on the fact that it isnt cliped even if the signal level
is below the noise (the limter has 60db of gain), the noise from the
preceeding stage may still drive the limiter stage into cliping, and the
point is that the device used isnt designed as a detector it is i beleive
designed more as an edge sensitive phase measurer, altough i havnt studied
it carefully myself. in fact if the limter isnt cliping then the phase
voltage output is no longer acurate, and certainly noise will introduce
considerable inacuracies.

Under these conditiones the only information is the position of the edge -
this is a smal part of the information contained in the original
signal/noise. its like fm and am, with a noisy signal on AM you can stil
make out what is playing, with FM however once its below the noise threshold
its all static you cant hear anything at all not even slightly, this is
becuase FM is usualy put through a cliping limter stage before detection.
you can get away with a quadrature comparison to find the phase.

it may be that reducing the drive to the detector may considerably improve
things so the noise does not dominate the edge position of the cliped
signal. this may also explain why 'improving' the amplifier makes things
worse.

Colin.

 

[Yannick]

its actualy quite easy, much easier than you would expect, and as the
circuit around the photodiode is so simple theres little room for problems
to arise. the only complication is generating an extra frequency that is
always 400khz diferent from your transmited signal, this would be easy with
a second DSS programed 400khz apart.

Yes but say your carrier is 20Mhz generated with a DDS. your frequency
superponated on the reverse bias of 200V will be 20.4 Mhz. Now what
about the phase between the two because i work with different
frequenties (carriers in this aproach) the 20Mhz from the 20.4 must be
in phase with the 20Mhz send frequency. otherwise you always measure
an other phase delay because the second dds will never load up at the
exact same time as the first.

i came up with it myself

very smart

yes, but if you look at the slope of gain versus voltage it is very steep
especialy in the region of maximum realistic gain, but this is exactly where
you want to use it anyway, this means a smal variation in bias voltage ~
500mv wil cuase a very large change in gain, this is more than ample to
provide for a very efective multiplier.

yes ofcourse, thats the part i still didnt got, i was still thinking
of mixing due the diode characteristic but you actually mix with the
voltage/gain characteristic.

the beauty is that if your signal is to strong you can reduce the bias
voltage and this reduces the gain of the photodiode and also the mixer gain,
making very efective agc. for smal change in bias voltage wich means minimal
change in apd capacitance.

yes but for a high dynamix range you sometimes have to reduce with
40-50Volts soo then the capacitance will change too much and the phase
delay will change significantly. Shouldn't it be better to reduce the
apmlitude of the AC signal on the reverse bias. Lets say this is 500mv
and if you lower it to 50mv this will give a 10 times lower signal at
the input of the amplifier but the capacitance will be the same (450mv
wont change much).

you can forget TIA becuase you get a single frequcncy out of your mixer
stage. a simple voltage gain stage with no feedback resistor exept for dc
biasing is then by far the best. in fact the signal realy is so strong with
an apd runing at high multiplication you dont want too much amplification.

soo just amplify the voltage over the bias resistor. I am thinking now
about non linearity due the exponential APD characteristic but for
small ac signal amplitude it wont give a problem.

400khz bandwidth would be far to much, in fact the use of a narowband filter
reduces the noise and the reduction of bandwidth of 30mhz down to 3 khz
would give a further 100 fold reduction in noise.

but if you work with a narrow band filter the phase change will be
rapid due any frequency error (drift).Isnt it better to use a filter
with low quality(Q factor) soo ok more noise but less phase change, a
compromise between the two must exist.

ah ic yes if you change the phase detector to a linear one then you might
wel find a huge improvement anyway.

yes i suppose but this way is easy

I lost a bit of interest when i found they are comercialy available and
havnt done anything to it for some time, however my initial goal was to try
and resolve with much greater acuracy over more moderate distances for
machining aplications and so i need to increase the carier frequcny from
30mhz to upto 1ghz wich cuases more of a chalenge, the idea of using the
beat frequcny from 2 lasers is something i was considering, and wicked
several stage frquency multipler to get the mixing signal.

its a shame it exist already.

I am going to write a whole chapter on this one (optoelectronic
mixing) for my book (thesis) because this is the way to go... to make
it work in practice in one month i dont gonne try, but next year on my
own i will try, i am really curious what this will give in resolution
when i step uy my carrier to 100Mhz orzo (with a better DDS)

its a challenging project , i am really happy i did it, its far more
difficult i ever could expect but thats good because i learned a lot.

thanks,

Yannick