What a Signal Conditioner Actually Does — and When You Genuinely Need One

A charge-mode piezoelectric accelerometer produces an electrical charge and nothing else. It needs a charge amplifier to turn that into a usable voltage, and until it gets one, the signal is so high-impedance that a few metres of ordinary cable will ruin it.

An IEPE accelerometer — the type most people now use, also sold as ICP®, CCLD, Deltatron® and IsoTron® depending on who made it — solves this by putting the amplifier inside the sensor, right at the crystal. That is a genuinely good idea: the fragile high-impedance signal never leaves the housing, and what travels down your cable is a low-impedance voltage that ordinary coax handles happily.

But that built-in amplifier needs power, and it has only two wires to work with — the same two wires carrying the signal out. So the power has to travel in along the same pair the signal travels out. That single constraint is the reason a constant-current source exists, and the reason almost everything else in this article behaves the way it does.

The conditioner pushes a constant current — typically 2 to 20 mA, with 4 mA a common default — up the signal wire. The sensor's internal amplifier draws what it needs and sits at some DC voltage, usually somewhere between 8 and 14 V. The vibration signal then rides on top of that DC level as a small AC wobble. A capacitor at the conditioner's output blocks the DC and passes the AC on to your analyser. How much wobble is allowed is set by the room either side of that bias: the supply's compliance voltage above it, and the amplifier's own minimum working voltage below it. Sit at 12 V on a 24 V compliance and there is roughly ten volts of room upwards and a similar amount downwards before the amplifier bottoms out, which is why an IEPE channel is commonly specified with an output swing of about ±5 V — and why the peak signal voltage matters as much as the cable length once we get to drive current.

That resting DC level is called the bias voltage, and it is the single most useful diagnostic in the whole measurement chain — because it tells you the state of the sensor and the cable before you have looked at a single spectrum:

Bias near 0 V — a short circuit. The cable is crushed, a connector is wet, or the sensor has failed short. Bias near the supply rail (typically 24–30 V) — an open circuit. A broken conductor, a connector not seated, or a sensor failed open. Bias in the normal band but drifting — moisture ingress, or a sensor being cooked beyond its temperature rating. Bias correct and stable — the chain is electrically healthy, and any problem you then see is a real measurement problem.

A conditioner that reports bias voltage per channel turns a twenty-minute fault hunt into a five-second look. On a route where you are moving one sensor between dozens of points, that check is worth more than any feature on the front panel.

This is the part product pages usually skip. If your DAQ already has IEPE inputs, you do not need a separate conditioner. The constant-current source is built in, switchable per channel, and that is the end of the discussion. Every PhonoVibe channel is IEPE-capable, so for a straightforward vibration measurement the answer is simply to switch it on.

A standalone conditioner earns its place in four specific situations, and it is worth being honest that these are the only four that matter:

Your DAQ has no IEPE. A general-purpose voltage card, an oscilloscope, a PLC analogue input, an older acquisition system — none of them will power an IEPE sensor, and without power the sensor outputs nothing. A conditioner sits between them and makes the sensor work.

You need gain before the cable run. A low-sensitivity sensor and a long cable in an electrically noisy plant is a bad combination. Amplifying at the conditioner, close to the sensor, improves what arrives at the far end. Amplifying at the analyser after the noise has been picked up does not.

You need filtering at the front end. A conditioner with a switchable high-pass removes the DC drift and settling ramp that will otherwise eat your dynamic range; a low-pass helps keep out-of-band energy from folding back into your measurement.

You need channel isolation. On large installations, ground loops between separately-earthed machines inject mains hum into everything. An isolated conditioner breaks that loop.

There is a real physical limit on how far you can run an IEPE sensor, and it is not about signal loss — it is about the amplifier's ability to drive the cable's capacitance at high frequency.

Coaxial cable has capacitance, typically around 100 pF per metre. The sensor's internal amplifier has to charge and discharge that capacitance every cycle, and the current available to do it is whatever is left over from the constant-current supply after the amplifier has taken its own bias current. Push the frequency up, or the cable length up, and you eventually run out of current. When you do, the signal does not vanish — it distorts, and it distorts at the top of your frequency range first.

The arithmetic is short enough to do on the back of a route sheet. The amplifier has to supply a current of 2π · f · C · V to swing V volts peak at frequency f into C farads of cable, so the highest frequency it can drive undistorted is f_max = (I_supply − I_bias) ÷ (2π · C · V_peak). The amplifier's own bias current is around 1 mA on most industrial IEPE parts, so a 4 mA supply leaves about 3 mA to drive the cable. Thirty metres of 100 pF/m coax is 3 nF: at 5 V peak that gives 3 mA ÷ (2π × 3 nF × 5 V) = 32 kHz of honest bandwidth, which is more than the sensor has. Fifty metres on the same settings still gives 19 kHz. That is where 'a few tens of metres at full bandwidth' comes from — it is not a rule of thumb, it is that division.

Two things in that formula are worth staring at. The ceiling falls with cable length, which everybody expects, and it falls just as hard with signal amplitude, which almost nobody does. Take the same 100 pF/m coax out to 100 m — 10 nF — and drop the drive to 2 mA, leaving 1 mA. At half a volt peak the chain still reaches 32 kHz and behaves perfectly. At five volts peak the ceiling is 3.2 kHz. Same cable, same current, same sensor: the ceiling moved by a factor of ten because the signal got bigger.

That is a nasty failure mode, because it looks like a measurement result rather than an equipment fault, and it only appears when the machine is doing something interesting. A bearing-envelope band at 5 kHz on the end of that 100 m cable at 2 mA is above the 3.2 kHz ceiling the moment the signal there reaches a few volts peak, so it distorts on exactly the readings you care about — while the 1× at 25 Hz, which needs a thousandth of that current, looks perfect throughout. Nothing in the data announces the problem. Raising the same 100 m run to 20 mA takes the ceiling to 60 kHz and the problem disappears.

The practical rule: more current buys more cable and more bandwidth. At 4 mA you are comfortable to a few tens of metres at full bandwidth. For long runs, raise the drive current toward 10–20 mA and check the sensor's rating first — and if you are running hundreds of metres, stop using IEPE for that channel and use a 4–20 mA loop-powered sensor designed for the distance.

Everything above is one circuit, so here it is as a circuit you can drive. The simulator puts a constant-current supply at one end, a length of 100 pF/m coax in the middle and an IEPE sensor at the far end, and plots the actual voltage on the centre conductor. The top plot is what a conditioner's bias monitor sees; the bottom plot is what your analyser sees after the DC blocking capacitor. Set a fault and watch the two disagree.

Work through the four states you will actually meet. Healthy: bias sits mid-rail, comfortably between the amplifier's own minimum working voltage and the supply's compliance, and the signal wobbles on top of it. Open circuit: the current source has nowhere to push its current, so the bias slams to the compliance voltage and the signal disappears. Short circuit: the bias collapses to near zero and the signal disappears again. Moisture ingress: bias drops low and wanders, which is the state people misdiagnose most often, because the sensor is still producing something.

Then produce the fault that the bias reading cannot catch. Take the starved preset: 200 m of coax, 2 mA of drive, a 3 V peak signal at 5 kHz. The bias is perfect. The bias monitor says the chain is healthy. And the waveform at the bottom is a sawtooth, because the amplifier cannot charge 20 nF fast enough to follow the rising half of the sine. Raise the drive current to 20 mA and it snaps back to a sine. That distortion is a measurement result to anyone reading the spectrum afterwards, and nothing in the data announces it.

Count the channels you need now and add the ones you will need within a year, because channel count is the thing people under-buy. Confirm the drive current is adequate for your longest cable at your highest frequency of interest. Insist on per-channel bias monitoring — it will pay for itself the first time a cable fails. Take switchable high-pass filtering if you measure anything that settles slowly. And if your DAQ already does IEPE and your cables are short, buy nothing: the box you need is the one already in front of you.

Most 'faulty sensor' calls we take are a supply or a cable, not a sensor. A conditioner that shows you bias voltage per channel converts that whole class of problem into a glance.

If you already own a DAQ with IEPE inputs, we will tell you so rather than sell you a box you do not need.