A stereo version of this was constructed using a NE5532AN dual version of the NE5534. The performance of the low noise amplifier was measured using Rightmark. Because this version was designed to explore the sensitivity at low levels the a higher gain was selected. R4 was made 820 and R5 15k to give a gain of about 19x (as opposed to 11). A 10mV p-p signal was injected into the input via an attenuator of 10k in series with a 12 ohm resistor to ground. From the junction of the 10k and 12ohm resistor the signal was taken via a 15nF capacitor in series with the input, to simulate the piezo source capacitance. This meant a signal of 12uVp-p (equiv to 4.2uVrms) was present at the input. You can see the effect of the source capacitance in the rising noise floor at low frequencies where the input noise current is developed across a rising impedance.
The 4.2uV reference tone corresponds to about -105dBu, coming in at -18dBFS on the plot. The noise floor is about -66dBFS, 48dB lower, -153dBu measured in 1Hz. Scaling this up to the 22kHz bandwidth means 43dB has to be added (=20log sqrt(22000)), so en is about -110dBu A weighted for this amplifier when presented with the piezo impedance (modelled as 15nF in this case).
hum
These tests suffered high hum levels relative to the noise floor – this is because the amplifier was mounted in a plastic box. The capacitive source impedance of 15nF is about 200k at 50Hz, and as a result it is easy to couple hum into the amplifier capacitively. You really should build this into a metal box connected to signal ground at the input if you plan to use this indoors at low levels. For most sound art applications signal levels are so high this is not a major problem, and outdoors hum is also not generally an issue. Most outdoors interference is high-frequency RF which finds a low impedance into the 15nF capacitance across it.
noise – terminate unused stereo inputs
Running in mono with a stereo version of this amplifier it is important to terminate the unused input. If the unused input is left open, the noise of the high source impedance unloaded by the piezo source impedance is amplified. This results in a noise floor some 30dB higher, which makes it hard to work with quiet signals if you are getting the unterminated noise in one channel of the headphones. A 0.1uF capacitor across the unused input terminals works fine – you shouldn’t just short it as there is DC on the input since the piezo disk source capacitance is used as an input coupling capacitor. Wire a 0.1uF capacitor across a dummy plug if you use separate input jacks for L and R piezo disks, and plug that into the unused input to quieten the unwanted racket.
Noise comparison of unterminated and terminated channels (left and right channels respectively)
At this level of sensitivity you have to be ready for some unusual second-order effects – this is the disk held by the signal cable at arms length. The low frequency noise is the noise of the muscles acting to hold the cable steady, and the slight residual vibration of the piezo disk held by the cable at one end. The second example is the disk also held by the cable sticking over the side of a chair but held steady by a weight on the cable. If you raise the gain you can hear that I should have turned my PC off for the lowest noise. Both these tracks were recorded on a HiMD NH700, Mic Lo-Sens 20, where 0dBFS corresponds to -29dBu.
Piezo disk cable held at arms length, disk in air supported by cable.
Piezo disk cable weighted on a chair, disk in air supported by cable. Record level same as above.
Comparing this version with the original bench model using the NE5534, it is slightly more noisy, the signal to noise ratio of the 1k tone is 49dB compared with 62dB (in a 1Hz noise BW). You have to allow for the fact that the gain of this version is 5dB higher, and the wanted signal is 4dB weaker, so although the wanted signal looks about the same level the signal to noise ratio would be expected to be 4dB worse, so the comparison comes to about 62dB against 53dB, about 10dB worse. I should probably have tested the performance of the NE5532 before soldering it in, along with some other assumptions.
Hi, thanks for all the data. I’m planing on making a plate reverb. Already have everything except the metal plate itself. I made some simple amplifiers but all are unbalanced and with 200khz loss.
Would you recommend making 2 balanced mono amplifiers, one with a tl072 and the other with NE5534?
Thanks for posting your work!
Guido from Argentina
Hi Guido, not quite sure what you mean with the 200kHz loss, do you mean the HF rolloff is 200kHz, should be OK for a reverb.
You will have lots of signal on a reverb, so a single-ended input will be fine if you ground the plate. If you need a balanced output to feed a long way into a balanced input then you can use a unity gain inverting amp from the op-amp output.
Although in theory you could make a wholly balanced system from the sensor outward by isolating both sides of the sensor, a piezo disk capacitance to ground isn’t balanced because of the way they are made, so IMO you’re better off using an unbalanced amp to match the impedance of the sensor and ground the brass disk side of the sensor, and then generating the balanced output from the unbalanced output if you need one.
The choice of opamp is up to you, either one would be good. If you are driving long cables the NE5534 with a 100R resistor at the sending end will probably be more stable than the TL072 and will eb a lower noise, but you’ll have so much signal the ultimate noise performance won’t be such an issue.
My preamplifiers for piezoelectric elements, they have balanced input and output. Powered by 48 volt phantom power.
Ultra low self noise version:
https://github.com/Supermagnum/piezo-balanced
Two NE5534 op amps: https://github.com/Supermagnum/double-gain
Phantom power supply, with mono headphone jack, powered by 9 volt battery:
https://github.com/Supermagnum/48power
Hello Richard,
From your RightMark screenshots I take the LF roll off of the Piezo+5532 combo is about 6dB/Oct (20dB/decade) below 100 Hz.
Is there a way to lower that roll off to, say, 10 Hz ?
Best Regards,
Eelco de Bode
Amsterdam-NL
This was about looking at the the noise performance, so I dropped R4 to get more gain to make it easier to see the noise and probably should have upped C1 to compensate. I also didn’t qualify what the input impedance of my soundcard (or its LF rolloff is) so I suspect the output cap C4 is low.
Hello Richard,
Thanks for making these designs and the discussion about them available online! I’ve done the most basic modification of the low-noise preamp to make it stereo (by duplicating the circuit, sharing the power, and omitting the duplicate filter capacitor) and I’d like to include the resulting schematic on my website (properly attributing the original design to you) alongside photos of a stripboard build. As I can’t find any licence for your designs, I’d like to ask your permission to reproduce it in this way.
Thanks!
Please go ahead, with the one stipulation that you have built and tested the schematic you publish 😉 I’m sure you have, but there are too many mods publish that haven’t been actually made…
Good luck!
Can you provide a link to your webite?
Hi,
Thank You for these informative posts. I want to build a piezeo contact mic for a digital stethoscope we are developing at IIT (Delhi, India). Since the commercial mics are very expensive I want to build my own following your tutorial for the low noise high end pre amp.
The frequency for interest is 0-100 Hz. Do you recommend any other OP Amp for better frequency response at lower frequencies?
The opamp is fine. C1 and C4 determine the LF limit. You can go down to DC if you use balanced power supplies, or define the LF rolloff more accurately using a servo amp.
There is a LOT of garbage at low frequencies – it’s not electrical noise, it is the difficulty of holding constant pressure on the piezo. As this example showed:
> The low frequency noise is the noise of the muscles acting to hold the cable steady, and the slight residual vibration of the piezo disk held by the cable at one end.
Try it as it is, first, I would say
you need to specify your requirements more accurately at the LF end 😉 These days you’d do that with DSP or in the computer processing the output. Low-frequency analogue filters are odious due to the capacitors needed below 20Hz.
If your application is specifically at LF then the input noise current of the opamp developed against the rising source impedance may lead you to prefer opamps with a higher En but a lower In. Tragically these tend to be jfet and mosfet inputs and both these technologies have a high 1/f noise. But in practice I wouldn’t worry about it. The mechanical noise due to not being able to hold the piezo still is the dominant noise, until you have solved that the opamps aren’t the issue.
Thank you for the reply. The piezeo would be mounted on an assembly and would be held in a hand against a stable surface. So mechanical noise would not be a concern I think.
On the LF end I need to record audio frequencies in 5-50 Hz even 5-20 HZ would work. I plan to feed the output into an ADC and process it using DSP. Since the output of the pre-amp is in uV would I need another amplifier to sample the signal through an ADC?
My device is portable, 5V is available through USB. So I cannot power the op amp using a balanced supply without a power converter.
“or define the LF rolloff more accurately using a servo amp”
I did not understand this statement can you calrify?
For 5-50 Hz prototype it: Put a resistor in series with the output high enough that won’t damage your ADC when the output goes to +/-4V 😉 I’d estimate 100k but it all depends on Zin of the ADC. Diodes to V+ of the ADC and to gnd in the same sort of thinking as D1 and D2 might be a good idea too. The rolloff of C1 and R4 is about 3Hz so leave it be
Try it out. See what your signals are like, maybe with a scope first 😉
In my tests 0dBFS was -29dBu which is about 0.08Vp-p. But I had hardly any signal going in. Trust me, your problem with piezo is often too much rather than too little . I used a higher gain
You need to get this on the bench with real signals, only then can you optimise it or see if you need to do something else. You will need to qualify the source capacitance of your piezo disks, mine was ~15nF, others may be lower. That and R1 makes a rolloff 3dB down at ~10Hz so you’d have to up R1 to 2.2M to get that down to 5Hz. The input offset current of the NE5534 is 400nA which will develop 0.8V(!) referred to the input so you don’t want to go making that really high like 10M which would offset the o/p by 4V, burning most of the output headroom. So if your disk motional capacitance is really low needing R1 much higher either find an opamp with lower input offset or use a FET opamp (which have hardly any input offset current) and accept a higher noise.
But get it on the bench. Oh and make the brass disk of the piezo ground, else hum will ruin your life. You should note that medical equipment (you say this is for a digital stethoscope) normally has stringent requirements for isolation from the AC line. Obviously this is battery powered but you may need to optoisolate the output, in that case it’s convenient that the output is on a pedestal of 4.5V.
Digikey has some ideas for you.
It’s worth highlighting before it comes into contact with somebody who isn’t in 100% health 😉
DC servos can help you define the LF more accurately. But they are troublesome to design and more one for grizzled old hands at analogue design. Fig 4 shows the priciples. You don’t need to go that far for a 5Hz LF limit.
Main thing is to get it on the bench. You learn analogue design in the lab. Though learning from Bob Pease is a good mentor.
Thank you for the reply. That optocoupler suggestion is great!
I am ordering the components to test the pre-amp out before using the ADC and optocoupler.
Before that I have two more questions:
In one of the answer, you said that the above circuit won’t work with 5V supply. Is it so for my case as well?
Would using a felt mat/ rubber mat on my table eliminate mechanical noise and noise due to the table? Although I am concerned about the ESD due felt/rubber, any suggestions on that?