Usually when testing power supplies, we measure voltage, current, regulation and ripple/noise. The regulation gives us an indication of the DC internal resistance of the supply, but what about the higher frequencies?
Measuring power supply AC impedance is not as difficult as it sounds, now that we have software for measuring loudspeaker impedance. For this project, I use LIMP, part of the ARTA software package. The principle is the same, see below.
The excitation signal drives the variable current source. This gives an AC current through R and Z. Since we know the voltages at both sides of R, and the value of R, we can calculate this current, I=(Uref-Uprobe)/R. Once we know the current, we find Z as Z=Uprobe/I. Inserting for I, we get the formula shown in the diagram.
When measuring a power supply, there is a DC component in series with Z. The generator has to be able to cope with that, and it can also be an advantage to be able to draw a DC current. A device that can be inserted between the computer soundcard and the PSU is shown here. Referring to the schematic, U1 and Q1/Q2 is the controlled current source. The static current is set by P1, and the excitation level by P2. The static current is needed because the PSU tester can only sink current, and not source it. The static current must be higher than the peak AC current.
An extra resistor is added in series with Q2 to limit the dissipation of the transistor. The S2055AF is a high voltage/high current device, but it can easily run too hot when testing high voltage supplies. Use a good heatsink, and limit the dissipation to 10-15W. My first test jig used an IRF830 MOSFET without series resistor, and the transistors died very fast. The S2055AF is much more reliable.
The internal PSU needs to supply about +/- 12 to 15V. A negative supply is needed to be able to set the static current all the way down to zero, but it doesn't need to be more than -5V. I used a transformer I had available. Regulated supplies are not necessary, and a 2x12V transformer would also work well.
The excitation input connects to the sound card output. Probe connects to sound card Left input, and Ref to the Right input (this can be set in LIMP, but Right as reference is the default). Set the value of the series resistor in Setup->Measurements. For other details in the operation of LIMP, please refer to the manual
Connect a resistor between J2 and J3 to limit the dissipation in Q2 to about 10W. For example, if you are testing a 300V supply at 100mA, you would want about 50-100V across Q2. A 2-2.5kohm resistor would be suitable. It is not a good idea to aim for a too low a static voltage across Q2, as it needs some headroom for the AC current.
Turn both pots fully anticlockwise and connect the PSU. Then turn up P1 to a static current suitable for the PSU under test. Then run LIMP, and advance the excitation level as high as possible without clipping either the sound card or the output current. This gives the best signal-to-noise ratio. It may be a good idea to connect a scope across the Probe output to make sure there is no clipping.
Several PSUs have been tested with the improved test jig. Here are some examples of the impedance of typical designs. Some PSUs have peaks at the line frequency and harmonics of it (50Hz, 100Hz...). This is due to ripple.
This PSU is a standard high voltage PSU of the simplest kind: transformer, rectifier, capacitor, resistor, capacitor.
This is part of the PSU for the SE 809 amplifier I made. It consists of the transformer from a Tektronix 502 scope, followed by a GZ34 rectifier, 20µF/10H/50µF, and an AC shunt regulator. Below is the impedance of the PSU without the AC shunt reg. Note the 7Hz resonance.
This is the impedance of the PSU for the 809 amp including the shunt regulator.
Here is the impedance of a tube voltage regulator. It uses three triode connected EL34s in parallel, and an E83F as error amplifier. It was originally an industrial PSU, capable of 250V @ 250mA. I have made a new chassis for it, added a choke, and made it variable. Now it covers 170-450V. Measurement below is done at 250V.
By hooking up an extra filter after the regulated PSU, we can easily see how the output impedance is affected. Here is the impedance when the PSU is followed by a 10H/80ohms choke and nothing more. Note that the choke is not behaving like a real choke above 200Hz or so.
All measurements below are done at 170V.
Adding 10µF after the choke.
Adding 30µF after the choke.
Adding 220R in series with the choke, cap is 30µF as above.
As above, with a 1200 ohms load added across the cap.
Adding 680µF after the choke, no series resistor.
Only 30µF across output, compared to no cap.
RC filter, 220 ohms, 30µF.
Alan Blumlein showed a clever way to make the PSU appear as a pure resistance. The patent, US 2,035,457 (or British patent no 421,546) describes how this is done using compensation networks for each reactance added to the PSU.
I built a prototype to conduct measurements on, as follows: The transformer is 2x433V, with about 70ohms DC resistance. This is followed by a AZ50 rectifier. According to the data sheets, this would give a DC resistance of about 250 ohms total. In series with this I added a 10H/85R choke. To compensate for this reactance, a 75µF cap in series with 330R was shunted across the supply. Then follows the actual filter cap, 30µF/600V. This reactance is nulled by a choke of about 3H in parallel with a resistor. 220 ohms seemed to fit. Then comes another choke, 8H/60R, followed by 235µF in series with 220 ohms to ground. This gives a PSU impedance that varies from 160 to 210 ohms across the 2Hz to 40kHz band.
VR tubes can be used as simple shunt regulators, or as voltage references. But they do have some sort of resonant behaviour that can be troublesome. Below is the impedance of a typical VR tube, the 0D3. Without bypass caps, it has a rising impedance from about 100Hz, with a peak at 20kHz.
VR tubes cannot usually be bypassed with very big caps, 0.1-0.2µF is normally the limit before the circuit turns into a relaxation oscillator. But 0.1µF is not effective at the midrange frequencies, it just moves the impedance peak to 3.6kHz instead.
By using a resistor in series with the cap, it is possible to achieve a very much better impedance characteristic. The curves below show how the impedance can be made to vary no more than 13 ohms across the audio band.
Different VR tubes in series can make things more complicated. Different tubes may have different impedance characteristics. Below are the impedance curves for a series connection of 0B2 and 0A2, IIRC 0B2 is the lower (grounded) tube. The tubes were used to regulate the screen grid of a 807 in a single ended amp. Impedances were measured with the circuit in operation.
Here is the impedance of the 0B2, measured at the connection point between the tubes.
Bjørn Kolbrek Sep 3rd, 2011