Introduction

Accelerometers are extremely useful for a number of applications, even for the hobbyist. For me, a good accelerometer would allow me to measure loudspeaker enclosure vibrations, loudspeaker cone acceleration (closed-loop feedback), design low-vibration surfaces for optics or vinyl record playback, among other things. But these applications place pretty demanding requirements on a sensor. Low-level vibrations, such as those of a loudspeaker enclosure, or normal surroundings require a really low-noise sensor. Conversely, the acceleration experienced by the cone of a loudspeaker can be extremely high (hundreds of g’s for a subwoofer), requiring very wide dynamic range. In recent years, MEMS accelerometers have become ubiquitous thanks to their use in portable electronics like mobile phones. But MEMS accelerometers are fundamentally limited by the small test mass they use. The low mass inevitably creates higher self noise, and limits low-frequency response. Furthermore, very few MEMS devices are linear to very high accelerations such as hundreds of g’s, outside of a few devices intended for use as airbag sensors.

Conversely, piezo accelerometers offer substantially improved performance across all the areas I care about: noise, dynamic range, and frequency response but it is difficult to find just a pure sensor. Normally, piezo accelerometers are sold as IEPE accelerometers (integrated electronics piezo-electric) which are fully packaged in a housing with additional electronics. These sensors can be hundreds of dollars however, which is more than I’m willing to spend on this type of project at the moment. This is why I got so excited when I found the Senther Technology 540A accelerometers at Digi-key.

Senther Technology 540A Accelerometer

OK, I have to get something out of the way first, I’m not a huge fan of the company name. Saying “Senther” sounds like I have a lisp. But with that aside, they really did make a cool product. Here are some key specs from the datasheet:

• Measurement Range: +/- 500 g

• Linearity: 0.5% (percentage of full scale)

• Sensitivity: 5 mV/g (typical)

• Noise Spectral Density: 4 ug/sqrt(Hz) at 1 kHz

• 3-dB Bandwidth: 0.5 Hz to 28 kHz

• Output Impedance: 500 Ohms

• Power Supply Range: 3 V to 30 V

• Power supply current: 600 uA to 1 mA

• Temperature Range: -55 to 150 degrees C

That’s quite a list of specs! The performance here is more than enough for the projects I have in mind, and the other specifications such as power supply range and current make this a really easy device to use.

Figure 1 displays frequency response of the sensor, normalized to 100 Hz. You can see the sensitivity increases as the frequency approaches the self-resonant point, which the datasheet gives as 50 kHz (typical). I really plan to use this more in the audio bandwidth, so the increase in sensitivity at very high frequencies isn’t concerning to me.

Figure 2 shows the noise spectral density curve for the sensor. There are two variants of this sensor: a low-noise version (540A) and a low-power version (540B). I have a feeling the difference between the two sensors is just the internal JFET used to buffer the piezo element. The lower-noise version likely uses a higher IDSS JFET. The datasheet gives a noise spectral density of 4ug/sqrt(Hz) at 1 kHz. We can see from the curve that this isn’t quite in the broadband region of the noise curve yet, and the noise will likely continue to fall above 1 kHz. Multiplying the noise spectral density by the sensitivity (5 mV/g) gives an output noise voltage of 20 nV/sqrt(Hz). This will be a useful number for us as we design the preamplifier circuit and select components. Ideally the preamplifier noise should be substantially less than the noise of the sensor itself.

Figure 1: 540A Accelerometer sensitivity deviation vs frequency. I’m not sure what the blue curve is, possibly the 540B low power version?

Preamplifier Design

The datasheet for the 540A gives some application on what kinds of “care and feeding” this device needs. Figure 3 is the absolute minimum circuitry needed to use the sensor. R1 and R2 provide a DC voltage to the reference pin, which also sets the DC output voltage of the sensor. C1 is a bypass capacitor for the power supply. T1 is a JFET current source that provides a bias current for the internal amplifier (likely also a JFET) of the sensor. I’m not a big fan of using a JFET as the output current source in this application. While it does offer simplicity, and sufficient output impedance, the JFET is operating at its drain-to-source saturation current (IDSS) which is not a tightly controlled parameter. For example, one manufacturer of the J201 (recommended for the 540A) gave a spread of IDSS values from 0.2 mA to 1 mA. My concern is that a lower-IDSS J201 wouldn’t properly bias the sensor, or give datasheet level noise performance. But I might be over-thinking things.

Figure 3: Basic circuit for the 540A accelerometer, providing a power supply, reference voltage, and a JFET current source to bias the internal amplifier (about 1 mA).

While the circuit in Figure 3 will allow for basic functionality of the circuit, the output signal level will still be only be 5 mV/g. When using the sensor to measure very small vibrations this signal will likely be swamped by the noise of whatever data acquisition system is being used. The answer is to add some gain to the circuit, and the 540A datasheet also has a suggested circuit for that (Figure 4) but it has a problem. The circuit in the datasheet, shown in Figure 4, is a pretty basic single-supply amplifier circuit AC-coupled to the 540A output through C2 and R4. The problem is that the op amp also has gain at DC. This will really limit the amount of gain that can be applied to the AC-signal of the 540A without saturating the amplifier at the upper power supply rail. The simplest solution would be to add a capacitor between R6 and ground, which would drop the amplifier’s gain to 1 at DC. Now the DC output voltage and the AC gain can be set independent of each other. However, that capacitor would need to be large to offer good low frequency performance, especially if R6 is made very small for good noise performance. There has to be a better way!

The schematic for the preamplifier is shown below in Figure 5. I wanted this board the be able to connect to the Analog Discovery 2 (AD2) interface PCB I made in the laser triangulation sensor project. This means it would need to run from a 24Vdc power supply and provide a pseudo-differential output for the AD2. Similar to the general-purpose photodetector project, I used a TL431 shunt regulator to generate a 20V supply to power the sensor. My thought here being it would help remove some of the noise on the 24Vdc supply and provide a much more stable bias voltage in the circuit.

Connector J1 goes to a “daughter board” for the 540A accelerometer and has connections for the 20V supply (VBIAS), ground, the sensor output signal, and the sensor reference voltage (VREF) which may be generated locally on the daughter board or on the preamplifier board, hence the jumper. Transistors Q1 and Q2 form a BJT feedback current source that provides the 1 mA necessary to bias the internal buffer JFET of the 540A accelerometer. The BJT feedback current source should provide much more consistent output currents across multiple builds because the output current is roughly VBE / R3 and transistor VBE varies much less than the IDSS of a JFET. The 2N4401 transistors used here were selected just because they’re good low-noise transistors to have on hand. If you were to select transistors specifically for this application, it might make sense to use something with a higher Early voltage for higher output impedance.

Figure 5: KiCad schematic for the preamplifier PCB. The PCB consists of a variable gain amplifier, bias current source, and shunt-regulated supply

Moving to the op amp portion of the circuit, U1B generates a low-impedance 5V reference voltage (VREF) for the circuit to operate from instead of ground. The input, feedback, and output of the preamplifier circuit are all referenced to VREF. The input signal is first AC-coupled through capacitor C2 and resistor R5 to the input of the gain op amp (U1A). Noticed in the schematic that I used a polarized electrolytic capacitor for this job. My thinking at the time was that this should be fine, because the output signal of the 540A accelerometer has roughly a 10V bias, and the preamplifier circuit has a 5V bias, allowing for sufficient DC bias across the capacitor. The problem is that the leakage current of that electrolytic capacitor is significant and creates a pretty substantial offset, especially when using the preamplifier in the highest gain. It also makes it really hard to test this preamplifier circuit without the accelerometer daughter PCB attached. It may seem like the obvious solution would be to use something like an X7R ceramic here and that is one possibility (distortion from AC-coupling capacitors is usually not an issue if they are sized properly). However, X7R capacitors are highly sensitive to vibrations, something I’d love to demonstrate on here in the future and since I plan to use this circuit to measure vibrations, it seems like it shouldn’t be vibration sensitive. That leaves tantalum or potentially film capacitors as the best option for C2 in the future.

U1A is configured as a non-inverting amplifier with gains of 2, 10, 20.1, and 196.7 V/V. This will give output signals of 10 mV/g, 50 mV/g, 100.5 mV/g, and 983 mV/g. I was trying to have a 1V/g output signal level in the highest gains, but this is close from the available resistor values. The different gains are selected using a DIP switch to connect R6, R7, R8, R9 to VREF or any parallel combination of them. The 1k Ohm feedback resistor R10 is shunted by an 8 nF capacitor (NP/0C0G dielectric to avoid producing distortion) to roll-off the gain at 19.8 kHz. R13 and R14 isolate any capacitance from the connecting cables between the preamplifier PCB and the AD2 interface and provide a 50 Ohm output termination.

The schematic shows an OPA1656 for U1A/B and while that is a fine choice for the op amp, I ended up using an OPA1612 when I assembled the PCB just because of its lower noise. My thought was that I could use this PCB for general low-noise amplifier duties, but for that to work I would really need to replace C2 with a less-leaky alternative. Furthhermore, since the DC resistances at each of U1A’s inputs are not match, it might be best to go with a FET-type (JFET or CMOS) op amp for very low input bias current. When used with the 540A accelerometer, the noise requirements for the op amp are not very stringent. A good design rule to follow is for the preamplifier input noise to be less than 1/3rd the noise of the sensor to avoid degrading the SNR of the system. That would suggest an op amp with an input voltage noise of 6nV/sqrt(Hz) or less in this application (540A output noise is 20nV/sqrt(Hz)). There are plenty of modern op amps that meet this requirement.

Figure 6: TINA-TI simulation schematic of preamplifier circuit and 540A accelerometer

Figure 6 shows the TINA-TI simulation schematic for the preamplifier. I made a basic model for the 540A accelerometer that would model the ref pin behavior, and the output impedance. Looking at it now, it’s very basic, and I could have done more modeling of the frequency response and sensor noise, but for the purposes of the preamp design. It’s fine. The frequency response of the preamplifier is shown in Figure 7. The 3-dB bandwidth in all gains is around 20 kHz as expected.

The input voltage noise spectral density of the preamplifier is shown in Figure 8. The noise is primarily from the 2 channels of the OPA1612 (1.1 nV/sqrt(Hz) input voltage noise) for all gains except 2 V/V when the noise from the feedback resistors degrades the noise further. The model for the 540A accelerometer and the input current source were removed for this simulation, otherwise the noise would be dominated by the 500 ohm resistor used to model the sensor’s output impedance. In any case, all of the curves are below the 6 nV/sqrt(Hz) number needed to preserve the SNR of the 540A accelerometer.

PCB Layout

I made two PCBs for this project: one for the preamplifier, and a “daughter” PCB just for the 540A accelerometer. The preamplifier PCB is shown in Figures 9 and 10. There’s not too much to say about the PCB other than it’s a 2-layer board for low cost, I did ground pours on the top and bottom and stitched them together with vias, and the ground pours are removed around the gain op amp’s inverting node to avoid stability issues that could come from parasitic capacitance.

Figure 9: Top side view of the PCB layout for the preamplifier board in KiCad

The daughter board allows for a cable to be connected to the 540A accelerometer and for the reference voltage to be generated local to the sensor if desired. It also has mounting holes to attach the whole PCB to a flat surface. A critical consideration for achieving datasheet performance of the 540A is a strong solder connection between the thermal pad under the sensor and the mounting PCB. For this reason I chose to attempt a DIY reflow in my kitchen using my stove and some very manual temperature control. The results were OK, but it did require me to go back with a soldering iron and touch up the four corners of the sensor to ensure a good electrical connection.

Testing

Once everything was assembled, my first tests were pretty simple: apply power and see if the green lights turn on, then wave the sensor around (like I just don’t care) and see if I can record an output signal on my Analog Discovery 2. With that out of the way (Figure 12), I wanted to see if I could measure the noise of the 540A accelerometer. To do this, I fixed the accelerometer at a 90 degree angle to the surface of my desk between two books and then set the Analog Discovery 2 to average the spectral density curve. According to the 540A datasheet it should have extremely limited response to vibration out of the z-axis, so the noise measured should just be of the sensor itself. For this measurement, the preamplifier was configured for a gain of 200V/V (1V/g output). Something that became apparent rather quickly when using the sensor is that the noise floor of the Analog Discover 2 is not very low itself and the 200V/V gain setting is the only one able to get the noise of the sensor above that of the AD2.

Looking at the noise spectral density curve in Figure 12, there are two curves, one with the accelerometer connected to the input and one with the input to the preamplifier shorted. Without the accelerometer, the input noise is about 6nV/sqrt(Hz) at 1 kHz. Since the preamplifier is in a gain of 200 V/V (actually 196.7 V/V) and we know it has an input voltage noise of 1.8nV/sqrt(Hz) at 1 kHz (Figure 8) we can back calculate the input noise of the AD2 to be 1.1 uV/sqrt(Hz) at 1 kHz…yikes. That’s really noisy! On the bright side, when the 540A is connected, the noise only increases to 14.4 nV/sqrt(Hz) which is substantially better than the 20 nV/sqrt(Hz) I was expecting from the datasheet values. However, this could also mean the sensor has lower sensitivity than stated in the datasheet and I wouldn’t really have any way to know. And that leads me to thinking about how to calibrate this sensor…One final thing to note, there seems to be some external interference causing a spike in noise around 7-8 kHz. The spike noticeably decreases in amplitude when the 540A is disconnected, which means the noise is likely being picked up by the unshielded cabling I’m using.

Closing Thoughts

I’m really excited about the possibilities for these accelerometers. They fill a gap in the market for higher performance at lower cost, which may also be why the 540A is sold out at Digikey at the moment. That being said, I definitely view this project as a “version 1.0” that will need some updating to really achieve the full performance of the sensor itself. The external EMI being picked-up by the cabling is frustrating and I’ll move to shielded cabling in the future. I also think it might be better to split the project into two boards differently: a dedicated accelerometer circuitry board and a general purpose low noise amplifier board. In the future I would also like to make a version of this project that interfaces with a USB soundcard that has a microphone preamplifier and phantom power (something like a Focusrite Scarlett 2i2) rather than the Analog Discovery 2. The soundcard would have a substantially lower noise floor, and would open up this project to a wider audience.

I’ll be updating this page in the future with more testing of the accelerometer and a calibration section, check back soon!