D. I. Y. Watch Timing Machine.

[wlysenko]

Sorry. The upload of my graph did not work. Trying again.

watchpreampgain.pdf

[svorkoetter]

8 hours ago, wlysenko said:

Sorry. The upload of my graph did not work. Trying again.

watchpreampgain.pdf

PDF is not a format supported by the forum, since it's not an image format.

[wlysenko]

Okay, here is a JPG 1024 pixels wide.

[svorkoetter]

That looks approximately as expected, although it's hard to tell with a linear gain axis. It should be down 12dB (i.e. down to 25%) at about 160Hz, and drop 24dB/octave (i.e. drop by a factor of 16 for each division by 2 of the frequency) below that. Likewise, it should be down 9dB (i.e. down to 35%) at about 11kHz, and drop 18Db/octave (i.e. drop by a factor of 8 for each doubling of the frequency) above that.

[rodabod]

There's a hell of a lot of gain there which is quite possibly necessary, but you are more liable to have oscillation problems as you go above 60dB of gain. 

[svorkoetter]

Yeah, the design has about 76dB gain in the middle of the passband, and it is necessary, since the signal from the piezo is very weak. I'm currently working on a non-watch-related application where I need even more gain, so I'm really going to have to work hard to avoid noise.

[wlysenko]

Here is the gain in dB. I account for the noise by using the corrected Vout = sqrt((Vout)^2 - (Vnoise)^2), where Vnoise=0.015 V.  The input voltage is 182 uV (using a 5490:1 voltage divider). I don't have 160 Hz and 11 kHz settings on my signal generator, but at 150 Hz, the gain is down 12 dB from the peak. At 10 kHz, it is down 18 dB. The latter does not agree with the prediction of -9 dB. What could be going on?

[svorkoetter]

I'm not sure what's happening here. To tell you the truth, I've never actually tested the frequency response of any of the amps I've built, although I have modeled them in a simulation tool. I'm about to head out on a week-long trip, but when I get back, I'll try to measure the response of the amp I'm currently using, which is built as shown on my site.

(I recently completed rebuilding a valve amplifier for my Hammond organ, at which time I put together the tools I needed to measure frequency response before and after the rebuild, so I'm now able to do this for the Watch-O-Scope amplifier as well.)

[rodabod]

Given that you don't need massive bandwidth, I'd consider putting an input transformer before the opamp stage. Something like 1:4 would give you 12dB of gain, though you'd want an opamp with very low current noise.  Also depends on piezo source impedance since a step-up transformer increases impedance by the square of the turns ratio. 

[robmack]

I'm making progress on my particular version of the Watch-O-Scope microphone amplifier.  So far, I've created and submitted a PCB design to Itead Studio for manufacture.  My PCB starts with Stefan's basic filter design, added the resistors and caps to the virtual grounds and made everything SMD.  I expect the boards will take some time to arrive in the post (about 1 month) but I'll begin construction once I receive them.  I've ordered extra boards so I would be willing to build additional units for those here who don't have the electronics skills; All depends on the performance of the design.

[wlysenko]

My latest version of the amplifier has only 3 mV of noise, compared to the original 22 mV. Probably good enough. I can barely hear the noise now. I made two changes. First, I changed to NE5534 op amps, which have much lower voltage noise. Second, I reduced the resistances in the first stage by a factor of 30 (and increased capacitances by the same factor) to reduce resistance noise. The diagram below shows the changes.

There is no bypassing other than what is show here, on the first stage only. For details see my noise article at http://wlysenko.blogspot.com.

 

[rodabod]

Did you ever measure the impedance of the transducer?

[wlysenko]

The resistance of my microphone (built just as in the Watch-O-Scope article) is greater than 500 Megohm and has a capacitance of 18 nF. Reactance is therefore 8.8 kohm at 1 kHz.

[svorkoetter]

Awesome work Walter! Now I want to redesign my preamp.

My one concern is the much lower input resistance. The reason I used such high resistances in the first stage was input impedance. With an 18nF (27nF in my case) transducer and a 1k input resistance, you effectively have a 9kHz high pass filter on the input.

However, perhaps that's a misunderstanding on my part caused by reading too many articles on using a piezo transducer for musical applications. In those cases, the input impedance is probably in the form of a resistor to ground, thus shunting the high frequencies to ground. In this case, the input impedance is a resistor to the op-amp input, so even the shunted signal is being amplified.

EDIT: As a first step, I may try simply replacing the first stage input resistors and capacitors, leaving the TL074 in-place. That should already make a significant improvement.

Edited October 15, 2016 by svorkoetter

[wlysenko]

Stefan (and everyone),

My noise predictions (calculations) are for a shorted input. So I compare to measurements for a shorted input. Of course, the actual input (piezo transducer) is different but, fortunately, the noise levels are less for the case with a microphone connected.  

For my latest version of the preamp, the noise measurements are as follows:
input shorted: 3.2 mV
mic connected: 2.2 mV
input open: 1.7 mV

For the original version of the preamp, the shorted and mic-connected cases were the same.

I haven't yet measured the frequency response of the latest preamp. I'll do that soon. Maybe you can provide guidance regarding changes needed to get the right frequency response for the low-resistance case. Or maybe I can learn filter theory (how hard can that be?). By the way, what should the passband be for this application?

[rodabod]

5 hours ago, svorkoetter said:

Awesome work Walter! Now I want to redesign my preamp.

My one concern is the much lower input resistance. The reason I used such high resistances in the first stage was input impedance. With an 18nF (27nF in my case) transducer and a 1k input resistance, you effectively have a 9kHz high pass filter on the input.

However, perhaps that's a misunderstanding on my part caused by reading too many articles on using a piezo transducer for musical applications. In those cases, the input impedance is probably in the form of a resistor to ground, thus shunting the high frequencies to ground. In this case, the input impedance is a resistor to the op-amp input, so even the shunted signal is being amplified.

EDIT: As a first step, I may try simply replacing the first stage input resistors and capacitors, leaving the TL074 in-place. That should already make a significant improvement.

This is why I asked about the transducer impedance. However, I was incorrect to ask: the opamp input impedance is still in the mega-ohm range as that 1K resistor is not strapped to ground; it's a series resistance. 

[svorkoetter]

1 hour ago, wlysenko said:

I haven't yet measured the frequency response of the latest preamp. I'll do that soon. Maybe you can provide guidance regarding changes needed to get the right frequency response for the low-resistance case. Or maybe I can learn filter theory (how hard can that be?). By the way, what should the passband be for this application?

I originally thought the passband should be about 160Hz to 12kHz, but I'm starting to think that most of the useful information lies between 800Hz and 8kHz, so a high-pass input at 800Hz should be fine.

The thing with R-C filters is that if C is fixed your hands are tied wrt R, as f = 1/(2*Pi*R*C). To get an 800Hz high-pass input, your input impedance needs to be 7.5k. Right now it's 1k.

[svorkoetter]

8 minutes ago, rodabod said:

This is why I asked about the transducer impedance. However, I was incorrect to ask: the opamp input impedance is still in the mega-ohm range as that 1K resistor is not strapped to ground; it's a series resistance. 

No, unfortunately the input impedance of an op-amp wired in a feedback configuration as shown is exactly equal to the input resistor. That 1K resistor is not strapped to actual ground, but it is strapped to virtual ground, because the op-amp will do whatever it can (through the feedback network) to keep the - input equal to the + input.

[rodabod]

Ah, in which case you effectively have a very large pad on the input as not very much voltage from your transducer will drop across that input resistor. 

Sorry for not noticing the opamp config - I've viewing on a phone so it's slightly hard to read. I'd set that input resistor to 10k plus and reduce the gain if need be. 

Edited October 15, 2016 by rodabod

[wlysenko]

22 hours ago, svorkoetter said:

I originally thought the passband should be about 160Hz to 12kHz, but I'm starting to think that most of the useful information lies between 800Hz and 8kHz, so a high-pass input at 800Hz should be fine.

The thing with R-C filters is that if C is fixed your hands are tied wrt R, as f = 1/(2*Pi*R*C). To get an 800Hz high-pass input, your input impedance needs to be 7.5k. Right now it's 1k.

Okay, I now understand a bit about filtering. I have two comments and a suggested modification. First, C1 has nothing to do with anything. It can be eliminated. With R1=1 k ohm (to keep resistor noise down), the RC value corresponding to the transducer (27 or 18 nF), does not correspond to 800 Hz so my modified preamp does not have the optimal passband. But it is not completely useless. I can listen to a watch ticking and not hear any noise.

Second, I have been measuring frequency response incorrectly. I should have put an 18 nF capacitor in series with my 180 uV, 1 ohm signal source to correctly characterize the microphone/amplifier system.

 Consider the following.

The low-resistance buffer provides a zero-impedance source to the first stage. Now we can use R1=1k and choose C1 to correspond to a frequency of 800 Hz, which is 0.20 uF. Unless someone points out why this won't work, I will try something like this.

[svorkoetter]

My op-amp-fu isn't strong enough to say if that will work or not, but it looks like it might. Certainly worth a try. The other alternative is to use a FET input stage, as has been suggested before in this thread (which at the time I rejected as unnecessary, since with my high resistance but noisy first stage, the input impedance was exactly what I wanted it to be).

Yes, I agree that C1 serves no purpose, theoretically. It does help to keep AC hum, picked up between the microphone and the circuit, out of the input though. A value of 0.33uF might work better in this case, without appreciably affecting anything else.

[rodabod]

The way I see it, you are going to get around 10% of your mic signal actually landing in the opamp, so you effectively have around a 20dB pad as 90% of the source voltage will dissipate in the transducer itself.

[Endeavor]

Even though I do understand the working of RC-filters, this discussion is way above my head. I'm very curious where this thread is leading to and can give only my hats off to all the efforts people put into this to make the W.O.S. working even better. I get now much better results with a wooden microphone stand (instead of a stand made out of nylon-plastic) and placing the watch-lugs (instead of the crown) agains the piezo (much stronger signal).

An amplifier which filters out the unwanted noises would be fantastic and I'm following this thread with great interest !!

Edited October 17, 2016 by Endeavor

[svorkoetter]

10 hours ago, rodabod said:

The way I see it, you are going to get around 10% of your mic signal actually landing in the opamp, so you effectively have around a 20dB pad as 90% of the source voltage will dissipate in the transducer itself.

Because the transducer is capacitive and the input is resistive, it's not a pad (audiophile-speak for attenuator), it's a filter. With the 18nF transducer and 1k resistor, the cut-off is at 8.8kHz. Above that frequency, all of the signal makes it to the op-amp. Below that, it drops off at 6dB per octave. So at ~800Hz, it is indeed 20dB of attenuation; at 8kHz, it's only about 1dB.

Edited October 17, 2016 by svorkoetter

[rodabod]

4 hours ago, svorkoetter said:

Because the transducer is capacitive and the input is resistive, it's not a pad (audiophile-speak for attenuator), it's a filter. With the 18nF transducer and 1k resistor, the cut-off is at 8.8kHz. Above that frequency, all of the signal makes it to the op-amp. Below that, it drops off at 6dB per octave. So at ~800Hz, it is indeed 20dB of attenuation; at 8kHz, it's only about 1dB.

Ahh!!!! Now I get you. My apologies; I didn't realise that you were using this to your advantage to essentially form the first filter stage. Fantastic.