Electronics > RF, Microwave, Ham Radio

DIY ISN project for twisted pair conducted emissions

(1/1)

nctnico:
DIY ISN
No, there is no L missing! An ISN is a device used to measure conducted emissions in communication / data lines. It is related to a LISN that is used for power supply lines. A good purpose / use case is power over ethernet (PoE) where you can have power supply ripple riding on the data lines.

As some may have noticed I have been working to get a new PoE product through EMC testing and that didn't go smoothly for various reasons (mostly to do with the test setup and surrounding equipment). This made me want to have better equipment to do EMC related testing in my own lab. One of the items I didn't have is an ISN. Nowadays LISNs can be bought rather cheaply (and the DIY one I have seems to work like a charm anyways). However an ISN is woefully expensive. Like 4k to 5k euro. Hmmm, can I built one? I found this schematic in the CISPR 22 for a 2 pair ISN:



There are also variants specific for 4 pairs and more complex ones that can deal with 1 to 4 pairs. But I choose to built this one for simplicity sake. The EUT is the side where the device under test (DUT) is connected. The AE side is where associated equipment is connected (IOW the outside world). The purpose of inductor L1 is to isolate/filter whatever comes from the outside world so it does not affect the conducted emissions measurement. This ISN has an attenuation of 9.5dB between the EUT/DUT and the output.

The most complicated part is making the transformer. I looked around for what is available where it comes to an ungapped core with high Al and HF ferrite. I decided to make my first version using an RM4 core using PC200 material from TDK. The Al value is around 550 but I could only fit 80 windings making the inductance lower than required. Based on previous experiences with winding equal windings, I cut the wires to length first and then wound the 4 windings in one go. It is way to easy to lose count and have one winding with too much or too little turns. When winding all windings at the same time, at least the number of turns is equal. However, don't start ordering RM4 cores yet... it turned out I'm not too happy with this transformer.

Another parameter for an ISN is longitudal conversion loss (LCL). This is a measure for the imbalance between the pairs and it is different for the type of cable (CAT3, CAT5 or CAT6). There is a formula in CISPR 22 that gives a conversion loss versus frequency. The resulting behaviour is in the graph below (taken from the spec sheet of the ENY81 ISN kit sold by R&S):



The longitudal conversion loss (LCL) for the ISN is set by the Zcat resistors in the schematic above. It looks like the commercial ISNs come with Zcat resistor kits. But how to calculate these resistors? And how to even measure it?

ITU to the resque. I found this principle measurement setup in ITU-T Recommendation O.9:



This should somehow lead to being able to determine values for Zcat. I transformed the setup to this schematic:



R1 and R2 are the line impedances, R3 is the load and Rx is Zcat.

A -65dB attenuation (lets follow R&S' terminology they use in the graph) means the signal on VL is 1/1778th of VI at 'DC' (IOW: at low frequencies).

Time for doing some math with the following input:
f is the attenuation factor (in case of -65dB it is 10^(65/20) = 1778)
Rpar is (R1 + R3) // R2 = 37.5 Ohm
Rser is (R1 + R3) = 100 Ohm

Rpar and Rser may not make sense at first sight but if you are going to calculate the currents through the circuit, these substitutions are handy to compress the formulas.

After some shuffling the terms around to express Rx as the input values I came up with the following formula:
Rx = (Rpar / Rser) * R3 * f - Rpar

For an LCL of -65dB (CAT5 cable) I get to Rx = 44.41k Ohm. I fitted the prototype with 47k Ohm resistors

Prototype
The prototype board ended up looking like this:


The only thing extra on the prototype are two MMBD101LT1G low capacitance diodes connected anti-parallel across the output to serve as protection / limiters for the attached spectrum analyser.

Measurements
For an ISN there are 4 parameters that are important:
- Longitudal conversion loss (LCL)
- Insertion loss versus frequency
- EUT / DUT to output frequency response (measured signal)
- AE to output frequency response (isolation from outside world)

Measuring the Longitudal conversion loss turned out to be a bridge too far. This is not simple and requires rather specialist hardware. I see no simple solution for this that would be accurate by design.

The rest is easely measured using my network analyser. The picture below shows the prototype in a casing and hooked up for AE to output measurement:



At the AE side I'm feeding all the wires from a 20dB attenuator. The ground from the input signal is bolted onto the casing. At the output I have a 6dB attenuator to prevent impedance mismatches at the ISN's output to mess too much with the cable impedance. The EUT side is terminated with 100 Ohm on each pair.

Graphs:
EUT to output:


AE to output isolation:


Insertion loss:


Conclusion
The measurements don't look too bad and a measurement with a real DUT shows a result that is close to that of an accredited EMC lab. However the AE to output graph shows there is resonance going on. The insertion loss graph also shows a typical resonance peak and I'd like the bandwidth to be a bit higher for ethernet. My assumption is that the relatively large number of windings is causing the transformer to have too much capacitive coupling. I'm planning on doing some tests with an RM8 core made from TDK T38 material which has an Al value of 12500. This means that 20 windings should be enough to reach 7mH. I also want to try and see if using wire with thick, low Er (Teflon) insulation makes a difference where it comes to capacitance between the windings.

T3sl4co1l:
I made one of these before, https://www.seventransistorlabs.com/Precompliance/index.html

The pot cores aren't optimal material (mu_r ~ 5k, best ferrites top out around 10-15k) but much better than what you got, so only a handful of turns were required.  The tapped inductors are suitable for Ethernet purposes, of course you would need a resistive or RC method like above if doing DC coupled signals.

Wind the pairs as, well, pairs; a split style bobbin could even be used to put them beside each other.  I suppose crosstalk shouldn't be much of an issue anyway, but probably loose wound all together takes a couple points off the maximum performance.  Maybe that's irrelevant for testing purposes.

I used quite fine wire to maximize turns per length, you'll need to use something bigger of course since you're handling DC power.

Tim

Navigation

[0] Message Index