How Can I Reduce Shot Noise in a Phototransistor Audio Application?

[MaximumPower]

Citizens of Physicsforums.

I would greatly appreciate your advice.

I've been wrestling with a pesky noisy phototransistor for quite a long time, and am at wit's end.

The sensor is being used to transuce an audio signal in the 1-5kHz range. Unfortunately the shot noise is introducing a broadband "hiss" into the audio that cannot be filtered, or at least not completely because it overlaps the audio signal.

Can anyone recommend a simple circuit to effectively "Dolby" filter such noise?

Thanks in advance for sharing your ideas!

Most obliged,

Jim

 

[Greg-ulate]

Could you average the signals from several phototransistors? If the noise is random it should average to zero from multiple detectors.

Could you modulate the source and attempt a lock-in method to remove the noise?

Welcome to the Physics Forums.

 

[yungman]

Before you worry all that, first, tell a little more about your design.
1) Is it you are sensing a light signal that modulated by an audio signal of 1 to 5 KHz.
2) Do you transmit in a pure dark environment or in environment with lights? If there is light from the environment, make sure the noise is not from the environment. One sure way is to totally block the photo transistor and see whether the hissing goes down.
3) If you are modulating the light, how hard are you modulating the signal...meaning what is the variation of the carrier? Make sure you modulate hard enough so you don't have to amplifier the photo transistor too much. Less gain means less noise.
Do you have any filter after the photo transistor pre amplifier? You are detecting from 1 to 5 KHz. Usually we put low pass filter at 5KHz to roll off any higher frequencies. That will make a difference.

If #2 fail, you need to show the circuit here to see whether you have too much noise gain. After all these, then you can worry about dolby and all.

 

[MaximumPower]

Thanks very much for your replies!

The signal I am transducing is surprisingly enough, a harmonica reed. Actually a set of ten reeds, summed. The noise appears to intensify with each additional sensor, rather than canelling out :-( But I'm intrigued by the idea of signal averaging... which I would know how to accomplish if it were a digital signal but cannot imagine how to do without some sort of sample/hold.

I'm relatively sure of the noise source as I experimented with a single isolated sensor with just a pull up resistor, illuminated by mini incandescent lamp. Completely sikent until the lamp is powered up.

The idea of modulating over a carrier occurred to me. .. . reflecting how some strain gages are (used to be) driven. But I do not have the expertise to design such a circuit.

I'll post the schematic tomorrow, but forewarn you that it is embarrassingly simple.

Thanks again for taking the time to share your ideas!

Jim

 

[Greg-ulate]

The noise appears to intensify with each additional sensor, rather than canelling out

How did you connect them in this case? To average the analog voltage I would buffer and Low Pass Filter each sensor and then average them with http://www.allaboutcircuits.com/vol_3/chpt_8/8.html.

For the modulation method, obtain a clock signal at an appropriate frequency, perhaps 150k-500kHz and use it to control a fast shutter that blocks your light source. Use this same clock to swap between inverting and not inverting your detector signal, and Low Pass Filter the output with a frequency roll off appropriate to your harmonica's range, 40kHz maybe? You could easily use this chip to do the job, as shown in the example labelled Lock-In Amplifier, replacing the first inverting amplifier with your buffered photodiode signal on pins 1, 16.

 

[MaximumPower]

Thanks again for your suggestions. Attached is the schematic, which I forewarned was rather simple. You'll see that there are already two stages of low pass filtering, but they don't seem to do the trick.

This circuit is not my design, per se. I actually paid someone to design it for me. But looking at it now more closely I find a number of features that don't look right.

1) The dropping resistors R4 and R5 seem to be too small. According to the sensor datasheet (attached), the collector current is nominally supposed to be 0.4mA, but the current value of 220 ohms suggests something like 100x that value.

2) the datasheets for the op amps state that the minimum bias voltage is 10V, but it looks like the Vcc in the current circuit is 8.15V

3) The feedback cap's are also not right. At first, I tried replacing them by computing the corner frequency. This improved the SNR, but barely. Then I tried empirically through trial and error, which was more effective. But still the problem is that the noise band overlaps the signal.

4) The LED driver current was set too high (appx 20mA) and as a result the transistors were saturating.

5) BTW, the frequency range from lowest note to highest note is: 274-2793Hz

I experimented with an external active filter, just to see if a sharp cutoff might do the trick. But even with a 32dB/octave filter, the noise leaks in.

Which is why I was fantasizing about a Dolby type filter ... it brings back memories of magnetic tape... it was impossible to filter out the hiss without throwing out the baby with the bathwater.

Now that's all off my chest, I'd like to digest what you write about the chopping and PLL approach. As I confessed earlier, its really beyond my expertise. But I would be pleased to pay someone to design a circuit if it would solve this abiding stubborn problem.

Thanks so much! I feel so much better now that I have someone to talk to!

Jim

ps: I would also appreciate your opinion about R4/R5 - will reducing the collector current help?

 

[sophiecentaur]

It isn't clear how you are modulating your audio onto the light. What are the details of the present transducer? There may be a link or attachment that I have missed? I can't find anything.

 

[MaximumPower]

I apologize if the attachments didn't get uploaded. I am trying again to attach them here.

I'm actually not modulating the light signal, per se. We're using SMT proximity sensors, with LED and phototransistor in the same package. The IR LED's are driven by a current source. And the reflection off the vibrating target (reed) is what creates the audio-frequency signal.

If necessary, I'd be happy with a generic sinusoidal signal of the proper frequency and amplitude - as long as it was free of noise (hiss).

thanks again

Jim

 

[sophiecentaur]

I was also interested in the physical setup. There will be a vast range in the modulation depth of the light coming off the vibrating object, depending on how it's actually illuminated and where you measure the light coming off. Basically, the Optics of the situation.
Also, looking for low level signals at low frequencies is not always the best. Synchronous detection, using a 'shuttering' frequency takes the signal to be detected up to a frequency well away from shot noise, hum etc. You don't need a phase locking system because you can use the same source for modulation and detection. The only difference with the equipment is the oscillator and the mixer. The filter then becomes a Band pass instead of a Low pass. I think you'll find this is pretty standard and used whenever possible / necessary.
However, I don't think you need be looking for a particularly low level signal in the first place if you make the most of the optical system in the first place. There is virtually no limit to how brightly you can illuminate your subject and how much you can vary the amplitude. This implies that you can get any signal to noise ratio that you need.

 

[NascentOxygen]

1) The dropping resistors R4 and R5 seem to be too small. According to the sensor datasheet (attached), the collector current is nominally supposed to be 0.4mA, but the current value of 220 ohms suggests something like 100x that value.

I expect the LED current was set to maximum in an effort to provide useable illumination of the reed, itself unlikely to be very reflective? Resistors R4 and R5 being the series 220Ω resistor between the 8.3V and 4.15V rails? It is difficult to see the typical I_C on your datasheet, quite a few details have been obliterated by your highlighter pen. Fig 3 in the datasheet shows the transistor is almost a constant current device, so use of a small R4 means the transistor is unlikely to saturate unless a large expanse of reflective white paper or a mirror is encountered.

2) the datasheets for the op amps state that the minimum bias voltage is 10V, but it looks like the Vcc in the current circuit is 8.15V

Most LM833 datasheet graphs do start at ±5V, though one shows a dotted line extending down to just below ±4V. The design does seem unnecessarily borderline, the designer could surely have made the 8.3V line 10.0V, though he must have been satisfied it was working okay at 8.3V,

4) The LED driver current was set too high (appx 20mA) and as a result the transistors were saturating.

I'd take some convincing to accept that is true.
http://img803.imageshack.us/img803/4666/holly1756.gif

 

[MaximumPower]

Thanks again for your thoughtful suggestions. The most encouraging and exciting think you wrote concerned the idea of chopping or heterodyning (or is it homodyning?) the signal. I understand this in principle, from my experimentation with Radio as a kid. But I never designed such a circuit. I would imagine that an IC must be available, like the AD630 you suggested earlier. However, I would also hope that there is a less expensive way of doing this with discrete components. (The AD630 is about $20, and I think I'd need two: one for each board - which is prohibitive.)

Regarding the illumination setup, I've attached a photo and diagram to illustrate the arrangement. The reeds are relatively reflective, made of polished brass/bronze.

I apologize for the yellow hilighting in the datasheet. I didnt realize that it made the text illegible. Attached is a clean version. The typical value of Ic is at the bottom of page 3. It refers to the "on state" inasmuch as the sensor was most probably intended for on/off binary detection of a target rather than the linear application I am forcing out of it.

Your recommendations concerning the intensity of illumination is, unfortunately, inconsistent with my observations. I found that the intensity of the noise is monotonically related to the intensity of the light (both when I use the built-in IR LED, or an independent grain-of-wheat incandescent.) The signal, on the other hand is unimodal. No kidding. The amplitude of the signal increases with LED current up to a point, but then drop off precipitously. If I move the target farther away from the sensor, then the point at which this drop occurs moves to a higher level of current. But still the same behavior is exhibited.

Illumination issues notwithstanding, I plan to implement your suggestion of lowpass filtering each of the sensors, rather than the entire bank all together. But I am most interested in implementing your frequency shifting/signal chopping/modulation/demodulation idea. Could I possibly hire you as a consultant to sketch the circuit? (I am able to lay out the board.) Or can you recommend someone who might be available and willing to do this? While I myself am eager to learn, I feel like I'd need to go back to school, and pay attention!

 

[sophiecentaur]

I now see the physical layout and it's obviously fairly cramped in there! Looking at the way you have mounted the sensors. I have some relevant points to make, the way the angle of the reed affects the illumination of the photo transistor could be increased significantly. The light flux hitting the sensor will hardly be affected by the distance from the reed in its motion - it will be due to the changing angle of the reed deflecting light away from the sensor to a varying degree. I think, if you had the sensor units rotated 90 degrees from their present orientation. Also, if the sensor were positioned at the end of the reed, the variation of angle would be at least double what you could expect if it's placed as in your diagram. This would give a much bigger variation in the amount of light actually reflected to it. I think. A big deflection would direct the light to one side, completely away from the line joining source and sensor. Did you experiment with this or was it just an arbitrary choice?
You say that the noise you get is apparently proportional to the light level. This must imply, I think, that the light level hitting the sensor is not changed a lot as the angle of the reed changes. The angle of the reed won't ever be great - perhaps a couple of degrees at that part of the reed and the trigonometry of the situation (and any curvature of the reed would be producing a reflection image with a position that varies only slightly. This could even imply that a narrow slot over both detector and source would produce would produce deeper modulation as the image would be deflected right away from the sensor. Needless to say, there would have to be a compromise for the slot size - too small and too little light would get through for the detector to register.
From your picture, it looks as if the mods I am suggesting could all be incorporated - albeit being a pain to do a rebuild.

 

[MaximumPower]

I cannot thank you enough for the time and effort you've been putting into this problem.

I'm very eager to implement your idea of chopping/modulating/demodulating the signal. Just wish I knew how. The AD630 is definitely an elegant one-stop-shopping chip, but a bit impractical for this application.

Regarding your suggestions on the light path etc., I do suspect that you are right about the light deflection due to the angular deflection of the target. On the other hand... the target is actually rather close to the sensor - within 2mm. And according to the Datasheet, the sensor is capable of sensing a flat object parallel to its face within that distance. (See attached Fig 1 from page 4.)

Of all the many assumptions and educated guesses that went into this design, the one meticulous thing I did was conduct a parametric experiment in which a single sensor was attached to a 3-axis stage. This permitted me to evaluate the sensitivity to x, y, and z positioning. I also did some angular sensitivity studies. I eventually configured the sensor layout so as to achieve the greatest amplitude signal but without buzzing against the reed (hence the reason they are not located at the tippy-tip.)

The illumination study that I did was less meticulous. I connected a single sensor in the most simple manner: with just a pull up resistor, and measured the voltage at the collector. Without exciting the LED, I illuminated the transistor with an old fashioned incandescent lamp, powered by a regulated supply. The result was white noise, audible as "hiss" when patched through an audio amplifier. Although I cannot say with certainty that it is shot noise, it is my best guess based on what I've read on various sources of noise, such as Johnson, Nyquist, dark, 1/f, etc...

That said, I do plan to try your experiment of masking the sensor with a slit. But the solution that brings me the greatest hope is your modulation/demodulation solution.

Speaking of which... I was wondering... what if... I were to transmit the signal using a wireless transmitter? I have one handy in the shop, although I think its based on FM. I've got nothing to lose.

Thanks again for your help.

Jim

 

[sophiecentaur]

I haven't had much time for this as the family are around but I suggest you look up topics around Synchronous Demodulation low signal measurement. At these frequencies, four quadrant multipliers will work easily and the shuttering frequency need only be a few tens of Hz to avoid any odd spectral foldback for your high notes. (There's some basic constraints when you want to modulate and demodulate).

I notice that these sensors are really meant for line following in robotics where the contrast is fairly high. I just wonder about how that distance sensitivity curve works and what exact optics is used for the light path - particularly with a semi-specular reflection that you will get off a metal reed definitely isn't a mirror. Using a slit may help - it could well be worth trying. The slit should be across the axis of the reed, I reckon and I think the sensor unit needs to be the other way round too.

 

[MaximumPower]

I will investigate your suggestions.
Please enjoy your family... this project has languished for years, it can wait a few more days. In the meantime, I'm learning about various noise reduction ICs... hoping to stumble on an inexpensive and simple solution: ala noisy signal in - clean signal out.
Thanks again.
Happy New Year!

 

[MaximumPower]

I think I might have found an IC that does what you are recommending.

AD7400A[/PLAIN] sigma-delta modulator


My hunt also led me down the path of dynamic noise reduction (DNR) and Dolby filtering, specifically a lineage of ICs originating with AD's SSM2000 circa 1996, which begat LM1894, which in turn is being replaced by TLV320AIC3256.

The TLV is only $1.96 on Digikey. It has all kinds of bells and whistles, including stereo audio amps. But I'm intimidated by its 40-pin package.

The LM1894 is on its way out, but still supported by TI. Its only $1.95, and features a much less imposing 14-pin SMT package.

Whenever you are back online, I'd appreciate your opinion.

 

[Mike_In_Plano]

I'd suggest trying a photodiode into a transconductance amplifier as an experiment. I've had nothing but pain from my experiences with phototransistors. Aside from having a large output signal, everything else about them is terrible.
For the transconductance amp, I'd use a low noise jfet or mosfet type op amp, such as the LMV751 with a +/-2.5 volt supply.

 

[sophiecentaur]

Dolby was originally invented (ifaik) as a way of reducing the audibility of tape noise for mini cassette recorders). It takes an audio signal and processes it so that channel noise which is subsequently added is less perceptible by boosting some frequencies on a variable basis. In your case, the noise is already on the audio signal so Dolby won't help, I'm afraid. the clever thing about the shuttering process is that it modulates the wanted signal to a part of the spectrum where there is less 'channel' noise. If the spectrum of your channel noise is, indeed, in the low frequency range then this will work. If it isn't, then there's nothing much you can do.
Of course, using the optimum design of amplification can also help (having the appropriate input impedance for your amplifier.)
I can see why you went for those sensors because they fit so nicely where they're needed but you may need to consider an alternative sensor. Did you consider some sort of capacitative pick up, for instance?

 

[jim hardy]

i may be wayyyy off track here

but had a thought...


Magnetic recording tape employs "Bias", a hf AC signal added to the signal to overcome tape's hysteresis and hiss. The technique was one of the War Prizes brought back from Germany at end of WW2.

Biasing is critical to this process of imposing the design of the symphony upon a medium which would tend to have random variations because of thermal energy and a kind of "inertia" in the form of hysteresis that resists the production of an undistorted image of the music

http://hyperphysics.phy-astr.gsu.edu/hbase/audio/bias.html

I'd tinker with the excitation current to the optocoupler to see if it helps. Perhaps add some ~40khz AC ? That you could do with existing circuit board...

Again - it's just a crazy idea at this point.
But i meant well.

old jim

 

[sophiecentaur]

Jim
Afaik, the bias just reduces hysteresis distortion which, in itself gives a gritty, programme related noise like the quantising 'noise' in ADCs, which is a distortion rather than random noise, really. Using a high bias frequency is a bit like extreme oversampling, which spreads the quantising noise over a wide bandwidth, mostly outside the audible range. The Recording Audio pre and de emphasis curves do a lot to help the basic SNR too because of the spectrum of random tape noise.
Mind you, Tape = Ghastly reproduction. It's a wonder we got away with it for so long.

I thought some more about a capacitative pick up and realized that the dreaded spit could ruin it with the small clearances needed for getting any sensible values of C. But I guess that would also be a problem with any optics relying on clean reflections.

 

[marcusl]

Just to reiterate sophiecentaur's insights: Your problem stems from having virtually no signal because of the very small angular motion of the reed and the nearly normal incidence angle that you chose for your light and sensor. It is no surprise that you see mostly noise.

It is always true in signal processing and instrumentation that 1 ounce of effort put into making the apparatus produce a better signal to noise ratio (SNR) is worth pounds of effort spent flailing about with modulations, filtration, adaptive noise cancellation, or whatever else you can come up with to put after your transducer.

Regarding some of the approaches currently under discussion: Dolby NR relies on the fact that tape noise has a colored spectrum. As a result, the audio signal can be predistorted, then signal+noise is filtered later. a) You think your noise is shot noise (a reasonable guess), which is white and thus not amenable to the Dolby approach, and b) you do not have the ability to predistort your signal, as already pointed out. As a result, this doesn't sound to me like a promising approach. Modulation (chopping) works to overcome 1/f noise, and is typically used in systems that operate at DC to maybe a few Hz where 1/f noise is problematic. You operate, instead, up to many kHz. Have you measured your noise power spectral density? Do you have reason to believe that your photo sensor 1/f noise predominates over the white noise? If it doesn't, then modulation won't help.

Before investing in partially thought-through efforts to "clean up" your almost nonexistent signal, I suggest that you spend some time now to rework your front end to boost that signal. At the very least, change the optical arrangement to be more sensitive to minute angular changes, as sophiecentaur suggested. Better yet, change to a capacitive pickup which, unlike your opto sensor, is directly sensitive to changes in distance between the pickup and reed. Changing to steel reeds and using magnetic guitar pickups might be another way to go. Or follow the example of Charlie Musselwhite and countless other blues musicians who hold a microphone to their harps. Just put a tiny microphone (electret or other type) into the harp case. You know this will work well!
In any case, maximize SNR first, and you're sure to have a better outcome.

 

[jim hardy]

You're probably quite right, S C.

Been trying to figure out how to modulate a carrier to get around that noisy optocoupler.
But that seems like trying to correct a bad microphone by using Armstrong's FM technique, which corrects noise in the atmosphere not in the mike.
My thought was this: if the signal were made into amplitude modulated carrier and diode detected exactly as in AM radio - what would it sound like? Could implement AGC to hadle that nonlnear opto curve.
But - enough idle rambling from me.

I'll watch this one with interest.


old jim

 

[sophiecentaur]

Just to reiterate sophiecentaur's insights: Your problem stems from having virtually no signal because of the very small angular motion of the reed and the nearly normal incidence angle that you chose for your light and sensor. It is no surprise that you see mostly noise.

... .. .

. . . . .

Before investing in partially thought-through efforts to "clean up" your almost nonexistent signal, I suggest that you spend some time now to rework your front end to boost the minute angular changes, as sophiecentaur suggested. Better yet, change to a capacitive pickup which, unlike your opto sensor, is directly sensitive to changes in distance between the pickup and reed. Changing to steel reeds and using magnetic guitar pickups might be another way to go. Or follow the example of Charlie Musselwhite and countless other blues musicians who hold a microphone to their harps. Just put a tiny microphone (electret or other type) into the harp case. You know this will work well!
In any case, maximize SNR first, and you're sure to have a better outcome.

It's good that we've at last got some other contributors to the thread. This really does need to be kicked around by several minds if we are to come up with an optimum solution.

Yes- a serious handicap.

They would need to be Stainless to resist the dreaded spit factor and many SS alloys are non magnetic. Also I believe 'they' like bronze reeds because of the timbre of the note. But spit would also affect the capacitance and even short out the reed - sensor gap.

I do think optical is the way to go but that initial off the shelf solution is hardly ideal. How about a fibre optic supply from a central LED source and then light diodes placed to give a very oblique incidence on the reeds to magnify the effect of deflection? However it's done, there will be significant non linearity.

A good reason for not using a microphone could be the wind noise / turbulence inside the body of the harp.

 

[sophiecentaur]

You're probably quite right, S C.

Been trying to figure out how to modulate a carrier to get around that noisy optocoupler.
But that seems like trying to correct a bad microphone by using Armstrong's FM technique, which corrects noise in the atmosphere not in the mike.
My thought was this: if the signal were made into amplitude modulated carrier and diode detected exactly as in AM radio - what would it sound like? Could implement AGC to hadle that nonlnear opto curve.
But - enough idle rambling from me.

I'll watch this one with interest.old jim

Jim
I think synchronous detection would be better than a diode detector as the SNR would be better. You would already have your local oscillator at the right frequency.

 

[jim hardy]

Thanks Sophie for the kind word. I am just interested, never did this so can't add much.

here's a scholarly paper by some folks who did measurements on harmonica reeds using proximity sensors in a lab setting.
Around pages 6-7 might be some data on displacement and frequency measurements that could be of interest to OP. Apparently there's more interaction between blow and draw reeds than one would expect.
http://www.www.harpinanawhinin.com/harp_bending_attributes_article.pdf

Their sensor was a mite big - i think 4mm,, ~KD-2400, Kaman Instruments Corp
http://www.intertechnology.com/Kaman/pdfs/KD2440.pdf

old jim

 

[marcusl]

They would need to be Stainless to resist the dreaded spit factor and many SS alloys are non magnetic.

400 series stainless steels (among others) are magnetic.

 

[sophiecentaur]

400 series stainless steels (among others) are magnetic.

True but I'm not sure the reeds would be massive enough to produce a big enough signal. Guitar pickups have only a low output power with strings which are hundreds of times more massive. In any case, would you want to produce a replacement set of reeds with DIY tools?

 

[sophiecentaur]

Thanks Sophie for the kind word. I am just interested, never did this so can't add much.

here's a scholarly paper by some folks who did measurements on harmonica reeds using proximity sensors in a lab setting.
Around pages 6-7 might be some data on displacement and frequency measurements that could be of interest to OP. Apparently there's more interaction between blow and draw reeds than one would expect.
http://www.www.harpinanawhinin.com/harp_bending_attributes_article.pdf

Their sensor was a mite big - i think 4mm,, ~KD-2400, Kaman Instruments Corp
http://www.intertechnology.com/Kaman/pdfs/KD2440.pdf

old jim

Hey Jim. That article was really interesting. I can manage to bend harmonica notes a bit and always wondered how it worked. I totally ignored the presence of the other reed in my attempted explanation. It always seems to be much easier on the 'suck' than the blow and the angle of the harp needs to be just right. I'll dig my harp out and see if the explanation helps with my technique (last practiced seriously when my 28 year old daughter was about 10).

btw, not only is that sensor a bit big but to fit two octave's worth of reeds out with them would be a bit pricy!

 

[johnbbahm]

I am not sure about the reflective characteristic of brass, but the 960 NM would
reflect great with aluminum.
Maybe you could sputter the reeds with a few microns of aluminum to improve the
IR reflectivity.
Also, what is the frequency nature of your hiss. can you use a scope to isolate the source?
Based on the size, I am assuming this is battery powered(cool Idea by the way).
The battery rules out power supply hiss.

 

[jim hardy]

I'm as bad as Detective Columbo - 'just one more thing...' MikeinPlano brought up the subject of amp induced noise.
The LM833 looks like a quiet amp but i wonder about the resistors surrounding it? Carbon is notoriously noisy i am told...

Mike - Just how noisy are phototransistors? Are diodes quieter?
This fellow observed noise differences due to manufacturing technique. He used a low noise preamp, unfortunately it was of vacuum tube type.
http://ufdcimages.uflib.ufl.edu/UF/00/09/77/13/00001/noiseinphototran00delarich.pdf

if irrelevant - please advise and i'll delete.

PS Sophie - glad the harmonica article was of interest . Thanks !

old jim

 

[AlephZero]

See http://www.harmsol.co.uk/ for how somebody else solved the problem (i.e. using a mcrophone!)

If you change the material of the reeds you are effectively making a new instrument from scratch. Even if you could get stainless steel reeds ready made, you would have to tune the instrument after you rebuilt it, and that isn't a trivial job!

Capacitance probes would probably work well. They have a frequency response up to 15 kHz which isn't quite "hi-fi audio" but might be near enough. The only problem is they would probably be too big to fit inside the case (but a line of cylindrical probes stickng out of the case might look "interesting"). The output can be as high as 10v/mm of movement, which should give you plenty of signal.

I would have thought you could fix any "wind noise" problems with a microphone easily enough by using a bit of acoustic wadding (the stuff that is used to fill loudspeaker cabinets to damp out resonances) as a windshield.

If you really want an optical solution, you could look at how optical computer mice work. The cost of the electronics in the cheapest ones must only be a dollar or two.

 

[johnbbahm]

I was looking at your picture again, maybe your noise is splash from the other
sources in the case.
How about a coat of black ink (flat) on all the brass surfaces except the reeds.
This would make the only reflective part, and the only moving part the same.
It also might help to look at this with a IR viewer (most digital cameras
can see some of the IR), and see how much IR is bouncing around inside your case.

 

[sophiecentaur]

See http://www.harmsol.co.uk/ for how somebody else solved the problem (i.e. using a mcrophone!)

If you change the material of the reeds you are effectively making a new instrument from scratch. Even if you could get stainless steel reeds ready made, you would have to tune the instrument after you rebuilt it, and that isn't a trivial job!

Capacitance probes would probably work well. They have a frequency response up to 15 kHz which isn't quite "hi-fi audio" but might be near enough. The only problem is they would probably be too big to fit inside the case (but a line of cylindrical probes stickng out of the case might look "interesting"). The output can be as high as 10v/mm of movement, which should give you plenty of signal.

I would have thought you could fix any "wind noise" problems with a microphone easily enough by using a bit of acoustic wadding (the stuff that is used to fill loudspeaker cabinets to damp out resonances) as a windshield.

If you really want an optical solution, you could look at how optical computer mice work. The cost of the electronics in the cheapest ones must only be a dollar or two.

The Capacitance Probes I can find all seem very bulky and they are supplied in their own case. I don't see why the frequency response is fundamentally limited - they just look for changes in RF signal level as the value of capacitance changes with movement. The commercial units are probably LP filtered for optimum noise performance. The problem with the capacitance method is that the value of capacitance of a reed against an Earth plane would be less than 0.01pF and the variation due to vibration would be even less. Also, no one has mentioned the linearity of any of these transducers which would affect the timbre of each note.

Your reference to optical mouse technology is interesting and it made me think about interferometry as a possible way into the problem.

@johnbbahm: I do like the idea of blacking the insides and dying the reed surface to improve wanted signal levels.

 

[MaximumPower]

EVERYONE,

Sincere THANKS for your continued discussion on this topic. I am embarrassed to say that I didnt notice the thread continued beyond Page 1. So I apologize for the radio silence.

Just to catch up, please allow me to respond en masse to your questions and comments:
====


Boosting SNR at the Source
Marcus: It is always true in signal processing and instrumentation that 1 ounce of effort put into making the apparatus produce a better signal to noise ratio (SNR) is worth pounds of effort spent flailing about with modulations, filtration, adaptive noise cancellation, or whatever else you can come up with to put after your transducer.

Jim: Truer words have not been spoken. Oh how I WISh I could eliminate the noise at the source! Hopeful that the illumination was the culprit, I’ve tried a variety of power sources and filters in the LED driver circuit. I also learned that LED’s are sometimes used explicitly for the purpose of *generating* white noise. But alas, I have found that even a dc-powered incandescent lamp produces the same amount of SNR.

Improvement of IR Reflectivity:
JohnBBahm: How about a coat of black ink (flat) on all the brass surfaces except the reeds. This would make the only reflective part, and the only moving part the same.
….I am not sure about the reflective characteristic of brass, but the 960 NM would reflect great with aluminum. Maybe you could sputter the reeds with a few microns of aluminum to improve the IR reflectivity.

SC: I do like the idea of blacking the insides and dying the reed surface to improve wanted signal levels.

Jim: I’ve experimented with coloring the reed with both a black sharpee marker, and white-out. (The sensor datasheet uses both a reflective aluminum target and a white piece of paper.) The effect is negligible. Yet the signal intensity itself is surprisingly good: on the order of 50mV without any amplification – with just a pull up resistor.

IR Viewer:
JohnBBahm: It also might help to look at this with a IR viewer (most digital cameras
can see some of the IR), and see how much IR is bouncing around inside your case.

Jim: I just discovered this handy tip recently. Its actually really cool. (Or hot, as the case may be.)

Dolby NR… not promising.
Jim: My original reference to “Dolby” was for loss of a better word. Now that I know what Dolby actually does, I think the term I’m searching for is dynamic noise reduction (DNR.) That said, I found several Dolby IC’s that do not require the recording to be encoded. I’ve been tempted to try them. But for the time being, I’m tinkering with the LM1894 which is a kind-of signal-intensity-controlled high-order-low-pass-filter.

Chopping
SC: Modulation (chopping) works to overcome 1/f noise, and is typically used in systems that operate at DC to maybe a few Hz where 1/f noise is problematic. You operate, instead, up to many kHz.

Excitation (related to above, I think)
OldJim: I'd tinker with the excitation current to the optocoupler to see if it helps. Perhaps add some ~40khz AC ? That you could do with existing circuit board...

Inductance Pickup
SC: Changing to steel reeds and using magnetic guitar pickups might be another way to go.

Jim: It probably is worthwhile. I tried building a miniature version of the Kaman eddy current sensor, with little ferrite rods wound with hair-thin copper. Couldn’t get a good signal from the high pitched reeds. Gave up.

Capacitive Probes:
SC: Did you consider some sort of capacitative pick up, for instance?

AlephZero: Capacitance probes would probably work well. They have a frequency response up to 15 kHz which isn't quite "hi-fi audio" but might be near enough. The only problem is they would probably be too big to fit inside the case (but a line of cylindrical probes stickng out of the case might look "interesting").

SC: The problem with the capacitance method is that the value of capacitance of a reed against an Earth plane would be less than 0.01pF and the variation due to vibration would be even less. Also, no one has mentioned the linearity of any of these transducers which would affect the timbre of each note.

Jim: I actually experimented with a type of poor-man’s capacitive pickup. Using nothing but the trace on a circuit board in proximity of the reed, I was able to frequency-modulate a kind-of broadband oscillator (made by Schmitt trigger, I think). The signal could then by any nearby FM radio receiver (tuned to virtually any station . I cannot remember why I dropped this idea. Perhaps fear of FCC.

Regarding Microphone:
Old Jim: Just put a tiny microphone (electret or other type) into the harp case. You know this will work well!

SC: Or follow the example of Charlie Musselwhite and countless other blues musicians who hold a microphone to their harps.

SC reply: A good reason for not using a microphone could be the wind noise / turbulence inside the body of the harp.

Jim: there are at least two reasons for avoiding a microphone. One is feedback. Harp players are constantly competing with electric guitar players. The second is somewhat philosophical. I consider this to be an entirely new instrument. Just as an electric guitar .NE. acoustic guitar + microphone, the electric harp allows far vaster range of effects. From Les Paul to Jimi Hendrix.

ps: A harmonica with built-in electret microphone was also previously marketed by a colleague. I think the name of the company is Harmonic Solutions.

Computer Mice
AlephZero: If you really want an optical solution, you could look at how optical computer mice work. The cost of the electronics in the cheapest ones must only be a dollar or two.

Jim: I also wondered, “how do the optocoupler and fiber optic people contend with this problem?” I think the answer is digitization. Alas, I’m an analog guy.

Armstrong
Old Jim: If the signal were made into amplitude modulated carrier and diode detected exactly as in AM radio - what would it sound like?

Jim: Although my formal training is in biomedical engineering, I got my start as a young boy exploring antique vacuum tube radios. I actually know much more about triodes and tetrodes than semiconductors. Which is both a handicap and a source of “out of the box” inspiration.

Noise Source, and Peripheral Noise (Amplifier, carbon, etc.)
Old Jim: The LM833 looks like a quiet amp but i wonder about the resistors surrounding it? Carbon is notoriously noisy i am told...

Jim: Although there are very few things about this problem that I understand for sure, I have unequivocally concluded that the source is the phototransistor. The definitive test was to simply attach one isolated sensor to a 9V battery through a dropping (pull up) resistor. I illuminated the transistor with an old fashioned incandescent (grain of wheat) lamp, again powered by a different battery. Result is HISSSSSS, loud as day. Light off: silence. Brighter the light: greater hiss. Up to a point of saturation (at least that’s what I call it) whereupon the signal cuts out completely.

Regarding illumination
SC: How about a fibre optic supply from a central LED source and then light diodes placed to give a very oblique incidence on the reeds to magnify the effect of deflection? However it's done, there will be significant non linearity.


Frequency Spectrum of the Hiss:
JohnBahm: Also, what is the frequency nature of your hiss. can you use a scope to isolate the source?

SC: Have you measured your noise power spectral density? Do you have reason to believe that your photo sensor 1/f noise predominates over the white noise? If it doesn't, then modulation won't help.

Jim: I do not have access to a spectrum analyzer, but as a next-best thing, I used a graphic equalizer to systematically try notching out the noise. Turns out that it is broad band. But from what I’ve read about audio hiss, the human ear finds certain frequencies more “irritating” than others. (namely 800-8kHz). The noise literally “sounds” like magnetic tape hiss, hence my initial inspiration for something “Dolby-like.”

Harmonica’s for Everyone
SC wrote: I'll dig my harp out and see if the explanation helps with my technique (last practiced seriously when my 28 year old daughter was about 10).
Jim: As a small token of my appreciateion, I would be pleased to send you a new harmonica. You can choose anyone you like from my website: www.turboharp.com.

Last but not least
the “Scholarly Article” discovered by Big Jim
Guess what. That’s MY article! Did you really think there would be two people in the world crazy enough to build a harmonica reed transducer??

I’ve actually been struggling with this project for many many years. Over 20 in fact. Attached is a little travelogue of the incarnations of the harmonica, beginning with the bulky Kaman eddy-current sensors. Progress has been slow, and in fits and starts. Since it is a total diversion from my day job, I’ve had to steal time from arguably more noble pursuits to work on this. Its been a Sisyphysian challenge.



Thanks again for sharing your experience, time, and creativity to help me with this challenge!

Jim

 

[MaximumPower]

Attachment - A brief travelogue of the electric harmonica."

 

[jim hardy]

Wow thanks for all that feedback !

And that article was YOU ! I am honored by the association.

I got my start as a young boy exploring antique vacuum tube radios.

i too started on vacuum AM radio.

You wrote:

Brighter the light: greater hiss.

aha - does the hiss vary with reflected light as well? If so , perhaps your "hiss" is the desired carrier.

Highpass to block audio out of hiss, amplify, halfwave rectify exactly as with old fashioned AM radio diode detector, lowpass hiss out of detected signal..

Trouble is you need one detector per reed.

But doubtless you're well ahead of me.


old jim

 

[humblepie]

I play harmonica, I tuned my own accordians, I can sympathise with the task at hand here and the seemingly strange behaviour of the noise from photodiodes.
Having worked on exactly these shot noise issues in research over the last 30 years, I hope my belated humble contribution offers some useful information.
When the incandescent light bulb experiment gave a strong hiss, there was a good (ie high) reading on the DMM.
But there seems something amiss here, and it's the absence of any modulation to generate an audio signal.
A comb or somesuch whipped in front of the light filament.
What your reed pick-up needs to see is modulation of the reflected light, not just lots of light per se.
Polish the ends of the reeds, shouldn't alter tuning much but take care by wedging with a slip of brass shim bridging between block sides and under reed. Then take the reflection tween LED and photodiode from that.
A narrow slit on LED and Sensor. Google Optical Film Soundtrack for the more advanced optical train, replacing the Variable Width soundtrack with your vibrating reeds. Yes, first experiments will be strange sounding. Maybe square waves, but the modulation will be there to get you into the acceptable S/N ratio region.

Useful link here - http://www.jensign.com//noise/noisecalculator.html

Dolby was originally invented to help Optical Sound on films where Pre-emphasis on quiet passages could be reduced on playback with a compander/expander process. This noise issue became important when amps and speakers became powerful enough to raise the overall theatre volume to high enough levels. Your reeds are already heavily pre-emphasised in this context.
I have no trouble picking up the reflection of a weak red pointer laser from a brass piezo sounder fed from the headphone socket of an iPod and getting good hifi sound. The movement of the reflecting surface must be in the micro-metre range.
Your reeds should be a cinch! You potentially have plenty of Contrast, Close Proximity tween LED/Sensor, Plenty of Modulation. The metal reed shield on a harp is a bonus too for avoiding mains hum issues.

By the way, I still haven't found ideal solutions to many problems where shot noise is an issue. The 'pumping' or 'breathing' effect of the noise following the amplitude of the light modulation is more obvious because it is 'single-sided'. It is correlated with the positive side of the audio wave. Well, yes, like a synthesiser special effect! I have been frustrated with how intrusive this can sound when there is clearly on the test bench a good S/N ratio! It needs a clear understanding of the physics of audio and perceived loudness at different frequencies to make sense of how to use photodiodes as audio sensors.

One issue to look forward to is Microphony, as in the harp acting as a mike. At that point you should be pleased to have the project successfully in hand. Knowing how reeds work, the limited dynamic range for example, should make for an easier project to me than it seems to have become. I have a chinese-made harp, LEDs and sensors to hand; so I'll give this a shot..
This noise is a strange phenomena at first encounter in DC conditions. Go to a movie or try to recollect the experience (35mm film analogue sound), enjoy it. Notice the soundtrack. Where's the shot noise from the photodiode?
Components; Phototransistors are often connected internally as a photo-darlington, with a higher dark noise. SMD versions of phototransistors and photodiodes are only pennies compared to the metal can and leaded legacy items which may be inferior specs, the price reflects different volume production. SMD is the way to go. To get better specs you will need better specs.

 

[humblepie]

I'll eat some humility on my suggestion to access the end of the reeds. I took a harp apart and it was obvious there isn't more than a sliver of metal reed and access is impossible compared to an accordion.

However, tidying the kitchen worktop of all the bits and bobs that tend to collect; led me to a serendipitious discovery which would allow the system and installation you have already worked on to succeed.

 

[meBigGuy]

Just a few observations: Maybe most of this is obvious.

1. If you are going to do this with "DC" light, filtering each reed's contribution before summing is important. Increasing the signal from each reed is also important.

2. Summing multiple sensors to reduce noise only helps if there is a common coherent signal. For example, multiple sensors looking at the same reed and phased properly. The coherent power sums, the noise as sqrt of sum of sq.

3. I don't understand the hetrodyning or delta-sigma suggestions. If the noise is already present in the signal it isn't going to help. Maybe I'm missing the obvious.

4. Sophie is correct about dolby. It really isn't "noise reduction" as much as "pre-equalization" before a noisy media.

6. A basic flaw with your design is that the noise of 10 sensors are summed together and always present. (this gets back to 1)

Maybe someone said this already, but I would modulate the LED current with a high frequency and then filter and detect that in the receiver.

Modulate the light at a frequency of 10-20x above the highest reed frequency component you want to detect.

 

[humblepie]

Is there any change in the reflected signal?

I can only admire the OP persevering with this idea and undertaking some practical experiments.

I have found that the reflection is great from the surface of a reed, but that it hardly alters the current in the diode when the reed angle changes as it would when being played?

Like I say, there are several ways to get around this using the original 'face on' approach. If the OP is still interested I'll explain further.

I suggest the OP does a 'static' DC test depressing and raising with a slip of paper under a reed and a DMM to confirm how much the current changes with the angle of the reed.
On this change hangs the whole issue of getting a resonable signal to noise ratio.
If you mount a single detector/sensor on a small piece of PCB, you may be able to temporarily mount it suspended at your original design height and position and still get access to a reed.

What is needed is more like a 'light-gate' - the sensors you are using are in a family of gross light change sensors. Like sensing the reflective spot on a turning gear wheel for example. either fully on, or fully off. Very bright ambient light is likely to be the only issue in such a case. Which is where 'high frequency modulation' techniques are relevant ie IR TV remote controls.

See my link to noise calculations above.
On this change in current value hangs the whole issue of getting a resonable signal to noise ratio.

In scientific instrumentation the modulation with a higher frequency is used to help remove 1/f or flicker noise. Shot noise is white noise. This should not be a marginal signal detection situation, which is where all the AD630 sychronous detectors etc are very helpful. Modulation adds it's own noise as well.

De-emphasis (low pass filter) may reduce the white noise to 'pink noise' - removing some of the hiss. As the frequency increases with the same amplitude, the energy will double for each octave. This is effectively a large amount of 'pre-emphasis'. However, the movement of each reed may be less when ascending the musical scale. It's difficult to tell whether this is the case because the size and thickness of the reeds in my sample get smaller as frequency increases. And I can't see what the reeds are doing while I play the instrument. A fixed set-up with foot-pump, a strobscope and video would be ideal - but it's really just another point to consider in terms of filtering the output.

The original stated frequency range is not going to give any kind of fidelity as far as the harmonics and timbre are concerned. From a quick check on an oscilloscope, a harmonica has significant harmonics going up to at least 5 kHz.

I'm interested to see how this project develops.
It's a fun way to look from a different viewpoint at some existing projects I'm trying to resolve.

 

[sophiecentaur]

SNR is what it's all bout.
What about an entirely different approach; did you ever consider a piezo sensor? The mechanical impedance could be good for coupling to a sensor screwed to the frame. There would be very little breath noise.
Just an idea.

 

[humblepie]

sophiecentaur - see his history of the electronic harmonica, he already tried the piezo contact mike approach.
Several decades ago the avant garde electronic musicians were fixing these sensors to all sorts of instruments and vibrating objects. They need the mass of a larger body vibrating with surface waves, as in a guitar body, or have force transfer through them like an 'under bridge' pickup. There are also electric reed organs which used the 'metal detector' principle and therefore the brass/bronze reed worked fine.
Thank you for posting the links OP;-
I have finally been able to download your Patent, History of your ELX Electronic Harmonica and your research paper on reed behaviour.

It would seem you now have all the information you need, but lack the inspiration after 20 years of frustrated development for a new lateral thinking approach.

Try a comb, a diode and a light bulb, get a feel for deep modulation versus spotlight DC illumination.
It's irrelevant that you maximised the output when aligning the sensors - if that maximum is a fraction of what should be happening.
It is a shame your research paper only has 'normalised' figures for reed movement. I'm guessing you never made any attempt to actually measure that factor because you were using the audio output from the sensors as the data to analyse?

 

[meBigGuy]

This may not be clearly described, but addresses the optics.

Maybe what is happening with the reflection off the reed is that:
1. The border of the LED "beam" is not sharp (distinct) enough, It is "blurry" where it hits the reed and then the sensor.
2. The reed angle change is not enough to cause the reflection to pass through the blurry border. In other words, there is not much change in the total amount of light falling on the photo-diode.

The sharpest the beam can be is if the led image is projected on the reed via a lens. Then if the motion of the reed is enough to cause the reflection to move on/off the photo-sensor, you get maximum possible amplitude. I don't know enough optics to know whether a simple ball lens will do the trick.

Another possiblity is to mess with the surface of the reed. If only a small section is reflective, then possibly you can position the sensor to get a larger change.

Or maybe you need to position the sensor to be off the very end of the reed where there might be a more distinct border.

But, I stand by my original observation:

Each channel needs to be filtered before summing. This is true for "DC" or modulated light. Understand that modulating the light is as simple as varying the series drive current. (although every channel then needs a filter and detector)

 

[sophiecentaur]

It's a longtime since this thread started and I forgot the piezo idea had already been mooted.

It's almost certain that the best optical system would use oblique illumination because this would achieve a long 'throw' * and hence more movement of the 'light spot' as the reed vibrates. Perhaps a cylindrical lens could focus the light from a linear light source or a set of optic fibres onto the sensors. The lens would give an improved illumination of the whole row of reeds and increase the SNR significantly, the light level arriving on the sensor could be modulated right down to zero for full amplitude vibrations if the spot could be made narrow enough to be deflected right away from the sensor.
An ideal optical system could involve a single moulded light 'manifold' with individual, spaced exits, with lenses of the right focal length to form an image on each sensor.
* There would be five times the deflection across the sensor if the light grazed the reed at an angle of 10°, for instance. The reeds would need to be well polished, though, for good specular reflections. Regular cleaning would be needed; there's no knowing what spit would do to the performance.