Fullscreen HTML5Med Res (??MB)HTML4

H igh-performance commercial power rails probes are available, but they cost many thousands of dollars, putting them out of reach for most hobbyists. This project proves it does not have to be that way. The probe described here offers good performance for less than $100 in parts. Passive oscilloscope probes are not really suited to looking at the ripple, switching noise and transients that can occur on power rails, especially those produced by switching converters. These are usually millivolt-level signals that are riding on top of a comparatively high DC voltage. Also, there is usually a lot of radiated noise that can be picked up by a standard passive probe with its 150mm-long ground clip wire. You will have to switch your oscilloscope to AC coupling to eliminate the DC offset to get the vertical resolution necessary to see the ripple and noise. This is fine if you are only interested in high-frequency artefacts, but no good for transients with time constants in the milliseconds range, like those you might encounter with a step change in load. AC-coupling introduces a highpass filter with a cutoff frequency in the 1-10Hz range into the signal path, which means the low-frequency and DC components will not be displayed accurately. You may be able to use DC coupling if your scope’s offset control has sufficient range. However, on the millivolt ranges, the offset is typically limited to a volt or two, so you likely won’t be able to get the trace on the screen at all. A power rail probe sits between the power rail being measured and the oscilloscope. It typically has an input impedance of 50kW, so it does not load the power rail too much, and an output designed to connect to a scope input in 50W impedance mode. The probe allows the DC offset to be removed but preserves the bandwidth from DC all the way to the upper bandwidth of the scope. P wer Rail Probe This is one of those pieces of test equipment that you don’t really need, until you do. It allows the measurement and evaluation of ripple, switching noise and transients riding on DC supply rails. Project by Andrew Levido If your scope does not have a 50Ω termination option, you could use a separate terminator (for example, the Amphenol 112667). Commercial models typically offer bandwidth up to the GHz range, can offset ±25V DC and can handle active signals in the range of ±1V. They usually attenuate the signal slightly, but this attenuation is known – and importantly – is constant across from DC to the upper bandwidth limit. The Power Rail Probe described here meets most of those specifications. It has a DC input impedance of around 50kW and can offset power rails up to ±25V. It can handle a signal amplitude of at least ±1V and has a nominal attenuation of -1.7dB ±0.3dB from DC to at least 100MHz. The actual limit is almost certainly a fair bit higher, but that is how high I can measure with my test equipment. The whole thing is built into a small plastic case using the same look and feel as the Current Probe and Differential Probe described in the January (siliconchip.au/Article/17605) and February 2025 issues (siliconchip.au/ Article/17721). Like them, it is powered by a lithium polymer cell that can be recharged via a USB-C power source. Design The block diagram of the Probe is shown in Fig.1(a). There are two signal paths in parallel: a low-frequency path, where the DC offset is applied, and a high-frequency path that keeps the shape of the waveform intact. This architecture is necessary because the offset circuitry uses op amps and is therefore limited in bandwidth. We will come back to the design of the offset circuitry later, but first we will focus on the two signal paths. If we assumed the offset is zero and the buffer is perfect, the circuit reduces to the LC parallel topology shown in Fig.1(b). I have added the 50W load presented by the oscilloscope and shown a voltage source for clarity. You will probably identify this circuit as a classic LC notch filter. At the Figs.1(a) - (c): the Power Rail Probe has a low-frequency signal path for the offset and a parallel high-frequency path, so it has very high bandwidth. The two paths would form an LC notch filter unless we introduce some resistance to lower the Q. siliconchip.com.au Australia's electronics magazine November 2025  47 (or DC), the capacitor will present an infinite impedance, and the inductor will present a zero impedance, so the attenuation will be given by the voltage divider formed by Rs and Rl. At very high frequencies, the inductor will appear to have infinite impedance and the capacitor zero impedance, so the attenuation will again be Rl ÷ (Rs + Rl). Series-parallel equivalent circuit The PCB is sparsely populated & none of the devices are overly difficult to solder. resonant frequency of 1 ÷ (2π√LC), the LC network will appear to have infinite impedance, so the response will have a distinct notch as shown in Fig.1(c). The steepness and depth of the notch is dictated by the Q of the filter. We obviously don’t want such a dip in our frequency response, so we need to introduce some resistance into the circuit to lower the Q. We can do this by inserting resistors Rs in series with the inductor and capacitor, as shown in Fig.2(a). I have added the ‘S’ subscript to the inductor and capacitor values for reasons that will soon become apparent. These resistors should be equal in value to keep the attenuation constant at the extremes. For example, at very low frequencies It is hard to visualise what happens at the resonant frequency, since the capacitor and inductor are no longer in parallel. However, we can take advantage of a special property of complex impedances, called series-parallel equivalency. This just states that at any given frequency, there is a parallel and series combination of elements that behave identically when viewed from the terminals. Fig.2(b) shows us what the network would look like after transformation. This means that there is a parallel Rp/ Cp circuit that behaves exactly the same as the series Rs/Cs circuit in the high-pass branch, and a parallel Rp/Lp circuit that behaves exactly the same as the series Rs/Ls circuit in the low frequency branch at the resonant frequency. The component values in the parallel circuits will differ from those in the series circuits, but the behaviour will be the same. The formula in that figure shows how the parallel and series impedances are related. With this transformation, the inductor and capacitor are now in parallel, so will have an infinite impedance at the resonant frequency. The impedance of the two paths at this frequency will therefore be determined by the two resistors in parallel (Rp). To keep the attenuation at the resonant frequency the same as at high Figs.2(a) & (b): the resistors in series with the inductor and capacitor form voltage dividers with the load resistance at very high and very low frequencies. This parallel equivalent circuit (b) behaves identically to (a) at any given frequency if the values are chosen appropriately. This allows us to calculate component values for a flat frequency response. 48 Silicon Chip Australia's electronics magazine and low frequency cases, we need each Rp to be twice the value of Rs. Substituting this relationship into the equation for Rp in Fig.2, we can see that Rs must be equal to Xc or Xl (which will be identical to each other at the resonant frequency). Using either one of these, plus the expression above for the resonant frequency, gives us the result that, for a flat response, Rs should be equal to √L ÷ C. I chose Rs to be 10W to give an attenuation of 1.2 (around -1.7dB), in line with the commercial units. This means the inductance should be 100 times the capacitance (in terms of henries and farads), so I chose 10µH and 100nF – both readily available values. These values give a crossover frequency of around 159kHz. The low-frequency path Fig.3 shows the full circuit of the Power Rail Probe. The ‘ground’ of the main signal path circuit (the horizontal line across the middle) is produced by op amp IC1c and the divider at its input. It settles at half the battery (cell) voltage, around 1.85V. The power supply for the op amps is therefore between ±2.1V and ±1.8V depending on the cell’s state of charge. Op amp IC1a forms an inverting, summing amplifier which adds the input voltage (via a 51kW resistor) with an offset voltage derived from potentiometer VR1. The input voltage is amplified by a factor of -1 (ie, inverted) due to the op amp’s feedback resistor also being 51kW. The ±1.8V present at the wiper of VR1 is amplified by -15.5, offsetting the input voltage by up to ±27V (or more if the battery voltage is higher). The second op amp, IC1d, is configured as an inverting buffer to flip the signal back to the right sense. I have used a potentiometer with a mechanical detent and centre tap that is connected to the virtual ground. This makes the zero-offset point very easy to find. This is helpful because it is very easy to lose the trace on the millivolt range if the pot can shift the voltage by ±25V. An easy-to-find zero point makes it much easier to get the trace back on screen. That said, a standard three-terminal pot would work just fine. The choice of op amps is quite important for the proper operation of the circuit. For once, we don’t care siliconchip.com.au Fig.3: the circuitry is fairly straightforward, using op amps to generate an adjustable offset voltage that’s applied to the low-frequency signal path. Potentiometer VR1 is a little unusual in that it has a centre tap and detent, to ensure that its wiper is at signal ground when centred. too much about input offset voltages, since the whole circuit is designed to add an offset. As long as it is no more than a few millivolts, the trace should be on the screen with the pot centred. We also don’t have to worry too much about the op amp’s input common-­mode range because we are using inverting amplifiers, which have their input voltages fixed at zero, and we have split supplies. We do need to use op amps that can operate at low supply voltages, and we need a reasonable output capability, since IC1d is driving a 60W load, and we’d like to swing as close to the ±1.8V rails as possible. We need the same drive capability for IC1c, as it is driving the other end of the same load. The most important op amp selection criteria is bandwidth, or more specifically, phase shift; a requirement not necessarily obvious given that the crossover frequency is only 159kHz. You could be forgiven for assuming that an op amp with a bandwidth of a few MHz would be fine in this application. Fig.4 shows the open loop gain and phase of one candidate, the TLV2460 family. These look promising at first, with a rail-to-rail output swing, ±80mA output drive, ±2mV offset voltage and a bandwidth of 6.4MHz. siliconchip.com.au However, close examination of the phase plot reveals a problem. Most op amps have internal dominant pole compensation that rolls off the open-loop gain response at -20dB/decade, as shown here. It also means the phase shift through the op amp is around -90° over much of its bandwidth. This roll-off is necessary for the stability of the op amp. If the phase shift were to reach -180° before the gain dropped below unity (0dB), the op amp would oscillate. You can see from the plot that the phase shift through the op amp starts to drop from -90° at around 300kHz, and is down to -100° around 1MHz. This will be a problem for us, since any deviation from -90° will cause a phase shift in our closed-loop response. If there is an appreciable phase shift through the low-frequency path relative to the high-frequency path, the two signals will add destructively, and we will see a dip in the overall frequency response near the Fig.4: the open-loop gain and phase plot for the TLV2460, from its data sheet, shows that the phase begins to deviate from -90° at around 300kHz, well below its gain bandwidth (GBW) figure of 6MHz. Australia's electronics magazine November 2025  49 Fig.5: a -90° open-loop phase shift (red trace to blue trace) results in a near-zero closed loop phase shift for a non-inverting amplifier. The phase shifts are exaggerated for clarity in this diagram. Fig.6: the open loop gain and phase plot for the TPH2504 shows that the phase remains very close to -90° all the way to 10MHz or thereabouts. The horizontal scale of this graph is strange, though. crossover frequency when both signals are contributing to the total. Op amp phase shift can be a bit hard to wrap your head around. How can an op amp with an open-loop -90° phase shift produce an amplifier with zero closed-loop phase shift (or 180° with an inverting amplifier)? Hopefully Fig.5 helps explain this. The upper chart shows the input and output voltage waveforms of an op amp configured as a non-inverting buffer. The red trace is the input voltage applied to the non-inverting input, and the blue trace is the output voltage, which is also applied to the inverting input via the feedback. I have shown an exaggerated phase shift between them to make the point. The green trace shows the difference between these waveforms. This is the voltage between the op amp’s two input pins that is amplified to produce the output. In reality, this voltage will be tiny, due to the high open loop gain of the op amp, but it will not be zero. You can clearly see that the phase shift between this open loop input voltage and the output voltage is close to -90° because of the dominant pole. If this phase shift were to increase (in the negative direction) to -100° like the TLV2461’s data suggests, the phase shift between the input voltage and the output voltage would increase to -10°. The TLV2460 is therefore going to introduce a significant phase error near to the crossover frequency, and we have two of these op amps in series, doubling the problem. The solution is to choose an op amp with a much higher bandwidth and/or a much more stable open-loop phase response, up to 10MHz at least. A bit of searching turned up the 50 Silicon Chip TPH2504 family. This is an op amp from 3-Peak – a company I had never heard of until this year. They seem to make some op amps with very impressive price/performance ratios. This one has ±2mV input offset, ±100mA drive capability and 120MHz gain-bandwidth (GBW). A quad pack IC of these costs less than $3.00 in small quantities. Fig.6 shows the open loop gain and phase plot from its data sheet. I have to say that this is one of the dumbest graphs I have seen in a while, because the horizontal scale increase by a factor of 100 every major division instead of by a decade like every other log-frequency graph you have ever seen. Why? Nevertheless, you can see that the phase shift remains near -90° all the way to 10MHz. extraneous switching noise into a device that only exists to allow us to measure the switching noise of the circuit under test! Fortunately, as the required signal amplitude is limited to ±1V, it is feasible to use the battery voltage directly, with the signal common derived from the mid-point as described above. This decision has two design implications. The previous designs used an unprotected LiPo cell and relied on the under-voltage lockout built into the DC-DC converter IC to prevent over-discharge. Not having this feature means choosing a cell with a built-in protection circuit (or adding a separate protection circuit, which would make the overall circuit more complex). I also chose to use a standard connector to provide a bit more flexibility regarding cell choice. Any cell with Capacitor and inductor the requisite protection board and a There is not much else to say about JST PH style connector that fits in the the signal paths. I used a 100V C0G/ case should work. NP0 ceramic capacitor in the high-­ The second design implication is frequency path because we want the that we now have separate ‘grounds’ capacitance to remain constant with for the signal circuit (half the cell volttemperature and DC bias. Don’t substi- age) and the charging circuit (cell negtute another dielectric like X7R here. ative). In most cases, the signal comIn the low-frequency path, I chose mon will be connected to mains Earth an inductor with a reasonably tight via the oscilloscope’s BNC terminal. ±5% tolerance and a fairly high It’s also possible (likely?) that the USB (40MHz) self-resonance. A typical charging port, and hence the charging inductor has a tolerance of ±20%, common, will be grounded. so ±5% is pretty good without being Unless we fully isolate the two cirunnecessarily expensive cuits, there is the potential for a short circuit. The solution is to use a twoPower supply pole power switch to ensure the two Unlike the Differential Probe and circuits can never be connected to the Current Probe, the Power Rail each other. Probe cannot use a DC-DC converter The charging circuit is identical to to create the power rails. The last thing my previous designs. The input is a we want to do is to inject a bunch of power-only USB-C connector followed Australia's electronics magazine siliconchip.com.au Parts List – Power Rail Probe Fig.7: assembly should be quite easy and fast as there are only a few parts. Take care with the orientation of the LEDs, TVS diode and the quad op amp. by a resettable fuse and a 5V TVS protection diode. These are included to protect against a rogue USB-C source applying a voltage higher than 5V to the connector. The two 5.1kW resistors signal the USB C power source to supply 5V at up to 3A. Yellow LED1 illuminates when the LiPo cell is charging and goes out when full charge is reached. The green LED (LED2) indicates that the unit is switched on. The charger, IC2, is configured to provide a 280mA charging current, so it should recharge a 400mAh cell in under two hours. The overall operating current consumption is 25-50mA depending on the signal level, so the battery life should be 8-16 hours. Construction All components mount on a single 56 × 82mm PCB coded P9058-1-C. For once, there are no tiny leadless parts, so assembly requires nothing but a soldering iron and a steady hand. You can commence by fitting the surface-­ mount parts according to the overlay diagram, Fig.7. Watch the polarity of the LEDs, the TVS diode and the TSSOP quad op amp. The rest don’t matter, or are hard to get wrong. siliconchip.com.au 1 double-sided PCB coded P9058-1-C, 56 × 82mm 1 front panel label, 41 × 60mm 1 Hammond 1593LBK plastic enclosure, 92 × 66mm 2 PCB-mounting right-angle female BNC connectors (CON1, CON2) [Molex 73100-0105] 1 USB-C power only socket (CON3) [Molex 217175-0001] 1 JST 2.0mm pitch 2-pin right-angle header (CON4) [JST S2B-PH-K-S] 1 10μH ±5% 480mA 240mW 40MHz SMD inductor, M4532/1812 size (L1) [Murata LQH43NH100J03L] 1 0.75A 24V M3226/1210 PTC polyfuse (PTC1) [Littelfuse 1210L075/24PR] 1 PCB-mount right-angle DPDT toggle switch with short actuator (S1) [E-Switch 200MDP1T2B2M6RE] 1 top-adjust, centre-tapped, centre-detent 50kW linear potentiometer (VR1) [Bourns PTT111-3220A-B503] 1 400mAh 38 × 25 × 6mm LiPo cell with JST PH connector (BAT1) [Core Electronics CE04375] 2 3mm diameter, 0.6in/15.24mm rigid convex light pipes [Dialight 515-1302-0600F] 1 knob (to suit VR1) 4 #4 × 6mm panhead self-tapping screws 1 small tube of cyanoacrylate glue (superglue) 1 38 × 25mm foam-cored double-sided tape pad 4 small self-adhesive rubber feet (optional) Semiconductors 1 TPH2504 quad 250MHz RRIO op amp, TSSOP-14 (IC1) 1 MAX1555EZK-T Li-ion battery charger, TSOT-23-5 (IC2) 1 yellow SMD LED, M2012/0805 size (LED1) 1 red SMD LED, M2012/0805 size (LED2) 1 SMBJ5.0CA unidirectional transient voltage suppressor, DO-214AA (TVS1) Capacitors (all 50V SMD X7R ceramic, M2012/0805 size, unless noted) 2 10μF 16V 1 100nF 100V NP0/C0G, M3216/1206 size 5 100nF Resistors (all SMD ±1%, M2012/0805 size, unless noted) 4 51kW 2 1kW 2 5.1kW 2 510W 1 3.3kW 2 10W The USB connector has surface-­ mount pads for the terminals, as well as through-hole mounting pads. The best way to mount this is to first solder it in place via the through-hole pads from the bottom, then turn the board over and solder SMT pads. Finish the PCB assembly with the battery connector, the BNC terminals, the switch and the pot. That’s all there is to it. Testing Check your work carefully, then connect the battery or an external supply set to 4.0V. Switch it on and you should see the green LED light. Use a multimeter to measure the power supply voltages with reference to one of the BNC connector shields. The bottom ends of the two 100nF Australia's electronics magazine capacitors just to the left of VR1 are convenient places to probe. You should read around +2V on the leftmost capacitor and -2V on the one to its right. Anything between ±1.8V and ±2.2V is fine. You can check the DC offset with the pot centred by measuring the voltage between the centre pin and shield of the output BNC connector. The voltage should be within ±5mV of zero. You can check the output voltage swing with the same set-up. Simply turn the pot either way until the output saturates. The voltages should be well above ±1.0V, even at the lowest battery voltage. With a fully charged battery, they will be closer to ±1.3V. You can check the battery charger is working by switching the unit off November 2025  51 Fig.8: a rendering of the finished assembly, with the battery plugged in and taped to the PCB, ready to install in the case. The test set-up used a conventional scope probe and a home-made RG316 probe to measure the output of this AC-DC converter module. and connecting a USB-C power source. Unless the battery is fully charged, the yellow LED should light, and the battery voltage should climb slowly. The LED extinguishes when the battery reaches full charge, at around 4.2V. Final assembly You can now fix the battery in place with a small piece of double-sided tape, as per Fig.8, then turn your attention to the case. Mark out and drill the two end plates and the top according to Fig.9. The aperture for the USB connector is best opened up after drilling by using a sharp craft knife or scalpel to remove the material between holes drilled at each end. I used a few small files to neaten things up. You can then apply the label to the top surface of the lid. The artwork is available to download (siliconchip.au/ Shop/11/2771). I printed mine full-size on glossy adhesive paper, then laminated that with some transparent self-adhesive vinyl. Cut it to size and fix into the recess in the lid, starting at one end to avoid capturing bubbles. I opened up the two light-pipe holes by pushing a sharp probe through the label into the holes in the case. The pot shaft opening is large enough to use a blade to remove the label over the aperture. Install the light pipes from the top of the case, and secure them on the underside with a drop of superglue. Thread the end panels onto the PCB assembly and lower it into the bottom of the case, making sure the end panels go into the slots provided for them. The board is held down by four 6mm-long #4 self-tapping screws. Pop the top case on and secure with the screws provided. I added four small self-adhesive rubber feet on to the bottom of the case. Fit the knob and you are finished. Using it The Power Rail Probe is dead easy to use. Connect the output to your oscilloscope with a 50W BNC cable and set the input to 50W termination. Set the vertical scale to a few hundred millivolts initially. Connect the Power Rail Probe’s input to your circuit and switch it on. With your circuit under test powered up, you should be able to adjust the offset pot to get the scope trace very close to zero. Fig.9: drill the top and the flat end-plates of the enclosure according to this diagram. The contoured end-plates supplied with the case are not used. 52 Silicon Chip Australia's electronics magazine siliconchip.com.au The finished Power Rail Probe, mounted in its 99 × 66mm plastic enclosure. The front panel label for the Power Rail Probe. Note the very small dots below the POWER and CHARGE labels. These are for the SMD LEDs, which shine through the panel via 3mm diameter light pipes. Punch 3mm holes centred on those dots after the label is affixed to the case. You can now zoom down the vertical scale appropriately, tweaking the offset pot slightly if necessary to keep the trace centred on the screen. Just remember the output on screen is attenuated by about 1.2 times (the attenuation will be within the range of 1.1-1.3 times). The connection you make between the power rail probe and your device under test will be the single most important factor in the measurement’s usefulness. Probing any high-frequency signal can be difficult, especially when you are working with switching power supplies. They tend to be environments rich in radiated and conducted interference that can easily upset your measurements. I set up a small experiment to demonstrate this, using a Zettler modular AC-to-DC converter rated at 15V and 5W. This is the one used in the Variable Speed Drive for Induction Motors project (November & December 2024; siliconchip.au/Series/430). I measured the unloaded output voltage of this switch-mode module with a conventional passive oscilloscope probe, with a 150mm ground clip and with a custom ‘probe’ made up of a short length of RG316 coax with a BNC connector fitted to one end. The photo at left shows the test setup. The scope capture (Screen 1) tells the story. The scope probe’s ground loop acts as a very effective antenna to pick up all sorts of switching hash radiating from the converter module. As a result, the underlying ripple is more-or-less invisible below the noise in the yellow trace from the probe. The home-made probe (green trace) has a INPUT OUTPUT 50 kΩ 50 Ω LOAD Maximum ±50 V Power Rail Probe POWER OFFSET CHARGE ±25 V Charge OFF - ON P9058 Charge much smaller loop area and picks up proportionately less noise. The faint vertical spikes you can see on the green trace are real signal artefacts caused by the very high voltage rates-of-change in the primary switch being capacitively coupled to the output. Their irregular spacing shows that the converter is operating in burst mode due to the very light load. Using coax probes like this is not something I invented. Commercial power rail probes come with similar unterminated cables for this purpose. However, there is a much cheaper alternative. I buy 1m RG316 BNC-toBNC cables from AliExpress and cut them in half to yield two test probes. You can reuse them many times, but they eventually get too short and have to be discarded. At the time of writing, three such cables cost less than $25.00 delivered. That’s way less than the cost of buying the cable and connectors to making them myself. Conclusion Scope 1: the results from the test shown at upper left. The waveform measured by the standard probe (yellow trace) is completely buried in switching noise, while the green waveform from the Power Rail Probe is much more informative. siliconchip.com.au Australia's electronics magazine A Power Rail Probe is far from the most essential piece of test equipment you will ever own. However, if you are looking at the dynamic response of converters that take place over tens of milliseconds, at voltage levels where you may run out of DC offset in your oscilloscope, there may be no alternative. Building this Probe is likely to be the most cost effective way to get that capability. SC November 2025  53